Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
University of Bath
PHD
Tilting Bundles and Toric Fano Varieties
Prabhu-Naik, Nathan
Award date:2015
Awarding institution:University of Bath
Link to publication
Alternative formatsIf you require this document in an alternative format, please contact:[email protected]
General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal ?
Take down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.
Download date: 29. Jul. 2021
Tilting Bundles And Toric Fano
Varietiessubmitted by
Nathan Prabhu-Naik
for the degree of Doctor of Philosophy
of the
University of Bath
Department of Mathematical Sciences
March 2015
COPYRIGHT
Attention is drawn to the fact that copyright of this thesis rests with the author. A copy
of this thesis has been supplied on condition that anyone who consults it is understood
to recognise that its copyright rests with the author and that they must not copy it or
use material from it except as permitted by law or with the consent of the author.
This thesis may be made available for consultation within the University Library and
may be photocopied or lent to other libraries for the purposes of consultation with
effect from . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Signed on behalf of the Faculty of Science . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ABSTRACT
This thesis constructs tilting bundles obtained from full strong exceptional collections of
line bundles on all smooth toric Fano fourfolds. The tilting bundles lead to a large class
of explicit Calabi-Yau-5 algebras, obtained as the corresponding rolled-up helix algebra.
We provide two different methods to show that a collection of line bundles is full, whilst
the strong exceptional condition is checked using the package QuiversToricVarieties for
the computer algebra system Macaulay2, written by the author. A database of the full
strong exceptional collections can also be found in this package.
ACKNOWLEDGEMENTS
Firstly, I would like to thank my Ph.D. supervisor, Alastair Craw. I could not have
hoped for a better supervisor; his enthusiasm and insight has been inspirational. He
has gone above and beyond to ensure that I make it to the end.
Much of my research began with the Macaulay2 quiver code written and generously
given by Greg Smith, for which I am grateful.
The Mathematics departments at both Bath and Glasgow are filled with fantastic
people to work alongside. I thank Alastair King and Markus Perling for their com-
ments on the thesis, as well as Tom Coates and David Calderbank for their help and
advice. In Bath and Glasgow I was lucky enough to share an office with delightful
people. I thank Jesus Tapia Amador for all of the conversations on toric geometry and
James Roberts for the laborious proofreading. Thanks to the friends and family who
have kept me going throughout the last four years.
Dedicated to Helen, who has been a source of constant support and happiness.
CONTENTS
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
1 Introduction 1
1.1 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Background 6
2.1 Toric Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Smooth Toric Fano Varieties . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Full Strong Exceptional Collections and Tilting Objects . . . . . . . . . . . 14
2.4 Helices and Calabi-Yau Algebras . . . . . . . . . . . . . . . . . . . . . . . 15
3 Strong Exceptional Collections on Smooth Toric Varieties 18
3.1 nnnvc-Cones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Cones Affected by Blow Ups . . . . . . . . . . . . . . . . . . . . . . . . . 20
4 Generation of Db(X) : The Frobenius Morphism (Method 1) 26
4.1 The Frobenius Morphism . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.2 Method 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5 Quiver Moduli and the Structure Sheaf of the Diagonal 31
5.1 Quivers of Sections and Moduli Spaces of Quiver Representations . . . . . 31
5.2 Quiver Moduli and the Structure Sheaf of the Diagonal . . . . . . . . . . . 34
5.3 Nef And Non-Nef Collections . . . . . . . . . . . . . . . . . . . . . . . . . 36
6 Generation of Db(X) : Resolution of O∆ (Method 2) 39
6.1 Resolution of O∆ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.2 The Toric Cell Complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.3 Method 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
7 Full Strong Exceptional Collections on Toric Varieties 53
7.1 Tilting Bundles Comprising of Line Bundles . . . . . . . . . . . . . . . . . 53
8 Future Directions 59
8.1 The Full Strong Exceptional Collection on D1 . . . . . . . . . . . . . . . . 60
i
CONTENTS
A QuiversToricVarieties: a package to construct quivers of sections on com-
plete toric varieties 63
A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
A.2 Overview of the Package . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
A.3 An Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
B Tables of Results and the Fourfold Contraction Diagram 68
B.1 Toric del Pezzo Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
B.2 Smooth Toric Fano Threefolds . . . . . . . . . . . . . . . . . . . . . . . . 69
B.3 Smooth Toric Fano Fourfolds . . . . . . . . . . . . . . . . . . . . . . . . . 70
C Further Examples 75
Bibliography 80
ii
LIST OF FIGURES
2.1 A lattice polytope in R2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 The dual polytope in NR and its corresponding fan . . . . . . . . . . . . 8
2.3 The fans in the blowup H1 → P2 . . . . . . . . . . . . . . . . . . . . . . 11
2.4 The effective cone and nef cone in Pic(H1) . . . . . . . . . . . . . . . . 13
2.5 The slice at height 1 of the fan for tot(ωH1) . . . . . . . . . . . . . . . . 17
3.1 The nnnvc-cones in Pic(H1) . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 The nnnvc-cones for the blowup H1 → P2 . . . . . . . . . . . . . . . . . 24
5.1 The quiver of sections of a full strong exceptional collection on H1 . . . 34
5.2 A quiver of sections on the smooth toric Fano fourfold E1 . . . . . . . . 34
6.1 A quiver of sections on tot(ωH1) . . . . . . . . . . . . . . . . . . . . . . 46
6.2 A quiver of sections on tot(ωX) . . . . . . . . . . . . . . . . . . . . . . . 48
6.3 A quiver of sections on the smooth toric Fano fourfold J1 . . . . . . . . 51
8.1 The set of relations J for the quiver on tot(ωD1) . . . . . . . . . . . . . 61
B.1 The torus-invariant divisorial contractions between the smooth toric
Fano fourfolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
iii
LIST OF TABLES
6.1 The additional arrows in a quiver of sections for tot(ωX) . . . . . . . . . 48
6.2 The arrows in a quiver of sections for the smooth toric Fano fourfold J1 50
8.1 The arrows in a quiver of sections for tot(ωD1) . . . . . . . . . . . . . . 61
B.1 Tilting bundles on smooth toric Fano surfaces . . . . . . . . . . . . . . . 69
B.2 Tilting bundles on smooth toric Fano threefolds . . . . . . . . . . . . . . 70
B.3 Tilting bundles on smooth toric Fano fourfolds . . . . . . . . . . . . . . 73
C.1 The arrows in a quiver of sections for the smooth toric Fano fourfold M1 76
C.2 Paths in the quiver associated to each torus-invariant representation in
Yθ ∼= M1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
v
CHAPTER
ONE
INTRODUCTION
Let X be a smooth variety over C and let Db(X) be the bounded derived category of co-
herent sheaves on X. A tilting object T ∈ Db(X) is an object such that Homi(T ,T ) = 0
for i 6= 0 and T generates Db(X). If such a T exists, then tilting theory provides an
equivalence of triangulated categories between Db(X) and the bounded derived cate-
gory Db(A) of finitely generated right modules over the algebra A = End(T ) via the
adjoint functors
Db(X) Db(A)
RHomX(T ,−)
(−)L
⊗A T
If X is also projective then one can use a full strong exceptional collection to obtain
a tilting object; a full strong exceptional collection of sheaves {Ei}i∈I defines a tilting
sheaf T :=⊕
i∈I Ei and conversely, the non-isomorphic summands in a tilting sheaf
determine a full strong exceptional collection. The classical example of a tilting sheaf
was provided by Beılinson [Beı78], who showed that O⊕O(1)⊕ . . .⊕O(n) is a tilting
bundle for Pn.
The combinatorial nature of toric varieties makes it feasible to check whether a
collection of line bundles on a smooth projective toric variety is full strong exceptional,
in which case one can construct the resulting endomorphism algebra explicitly. Smooth
toric Fano varieties are of particular interest; there are a finite number of these varieties
in each dimension and they have been classified in dimension 3 by Watanabe–Watanabe
and Batyrev [WW82, Bat82b], dimension 4 by Batyrev and Sato [Bat99, Sat00], di-
mension 5 by Kreuzer–Nill [KN09], whilst Øbro [Øb07] provided a general classification
algorithm. King [Kin97] has exhibited full strong exceptional collections of line bun-
dles for the 5 smooth toric Fano surfaces, and by building on work by Bondal [Bon06],
Costa–Miro-Roig [CMR04] and Bernardi–Tirabassi [BT09], Uehara [Ueh14] provided
full strong exceptional collections of line bundles for the 18 smooth toric Fano three-
folds. The main theorem of this thesis is as follows:
1
1.1. Structure of the Thesis
Theorem 7.4. Let X be one of the 124 smooth toric Fano fourfolds. Then one
can explicitly construct a full strong exceptional collection of line bundles on X, a
database of which is contained in the computer package QuiversToricVarieties [PN15a]
for Macaulay2 [GS].
In addition to low-dimensional smooth toric Fano varieties, other classes of toric vari-
eties have been shown to have full strong exceptional collections of line bundles – for
example, see [CMR04, DLM09, LM11]. Kawamata [Kaw06] showed that every smooth
toric Deligne-Mumford stack has a full exceptional collection of sheaves, but we note
that these collections are not shown to be strong, nor do they consist of bundles. It is
important to note that the existence of full strong exceptional collections of line bun-
dles is rare; Hille–Perling [HP06] constructed smooth toric surfaces that do not have
such collections. Even when only considering smooth toric Fano varieties, there exist
examples in dimensions ≥ 419 that do not have full strong exceptional collections of
line bundles, as demonstrated by Efimov [Efi10].
The tilting bundle we construct on each smooth toric Fano variety determines a
tilting bundle on the total space of the canonical bundle ωX :
Theorem 7.7. Let X be an n-dimensional smooth toric Fano variety for n ≤ 4,
L = {L0, . . . , Lr} be the full strong exceptional collection on X from the database and
π : Y := tot(ωX)→ X be the bundle map. Then Y has a tilting bundle that decomposes
as a sum of line bundles, given by⊕r
i=0 π∗(Li).
1.1 Structure of the Thesis
This thesis comprises of 8 chapters and 3 appendices. It is the combination of two
papers [PN15b, PN15a] by the author, with additional explanations and examples.
Chapter 2
Chapter 2 is divided into four sections and recalls most of the background material
needed. In the first section, we introduce the construction of toric varieties via the
theory of lattice polytopes. A fan Σ associated to a lattice polytope gives a combinato-
rial model for the toric variety and can be fully described by primitive collections and
relations (see Definition 2.4). We also consider the blowup X0 → X1 between two toric
varieties, the corresponding change to the fan for X1 and the resulting map of Picard
lattices γ : Pic(X0)→ Pic(X1).
The second section recalls some properties of line bundles on toric varieties, as well
as Batyrev’s classification of the smooth toric Fano fourfolds [Bat99]. A description of
the combinatorial change to the fan of a toric variety after a blowup (see (2.1.6)) was
used by Sato [Sat00] to complete the classification of the smooth toric Fano fourfolds; we
call the maximal smooth toric Fano varieties with regard to these blowups birationally
maximal. The map γ resulting from a blowup becomes important when we consider
how to produce full strong exceptional collections of line bundles on all smooth toric
Fano fourfolds from collections on the birationally maximal examples.
The third section of Chapter 2 introduces the reader to the definitions of a full
strong exceptional collection and a tilting object for the bounded derived category
2
1.1. Structure of the Thesis
Db(X) of coherent sheaves on a smooth variety X. We recall that the existence of a
tilting object implies that Db(X) is equivalent to the bounded derived category of a
module category (see (2.3.1)), and that tilting objects on varieties Y and Z determine
a tilting object on the product Y × Z (see Lemma 2.24).
In the final section, we recall Ginzburg’s definition of a Calabi-Yau algebra [Gin06],
as well as Bridgeland–Stern’s definition of a geometric helix [BS10]. The condition for
a helix to be geometric is central to the proof of Theorem 7.7.
Chapter 3
Chapter 3 focuses on how we can show that a given collection of line bundles on a toric
variety X is strong exceptional, by utilising the construction of the not-necessarily
non-vanishing cohomology cones (nnnvc-cones) in the Picard lattice for X as intro-
duced by Eisenbud–Mustata–Stillman [EMS00]. The strong exceptional condition then
becomes a computational exercise, which has been implemented into QuiversToricVa-
rieties [PN15a]. In the second section of Chapter 3, we show that these nnnvc-cones
behave well under the map γ : Pic(X0)→ Pic(X1) corresponding to a blowupX0 → X1.
This simplifies the process of finding a strong exceptional collection on a smooth toric
Fano variety such that the collection is the image under γ of a strong exceptional col-
lection on a birationally maximal smooth toric Fano variety – see Propositions 3.13
and 3.14 for more details.
Chapter 4
The procedure to check whether a given strong exceptional collection L = {L0, L1, . . . ,
Lr} on X generates Db(X) is less straightforward. We use one of two methods to show
that L is full, the first of which is introduced in Chapter 4 and is similar to the method
used by Uehara for the toric Fano threefolds [Ueh14]. This approach uses the Frobenius
morphism Fm : X → X, where m is some fixed positive integer; Thomsen [Tho00] has
shown that the Frobenius pushforward (Fm)∗(L) of a line bundle L on a toric variety
splits into a direct sum of line bundles. We use the Frobenius pushforward to obtain a
set of line bundles that are known to generate Db(X) and then show that L generates
this set by using exact sequences of line bundles.
Chapter 5
The second method we use to show that L is full uses the line bundles in L to obtain
a resolution of O∆, the structure sheaf of the diagonal embedding of X into X × X.
Chapter 5 and Chapter 6 focus on how we produce this resolution. Chapter 5 begins
by recalling how to construct a quiver of sections Q and the moduli space of quiver
representationsMθ(Q,J) corresponding to L, for some stability parameter θ and ideal
of relations J . We introduce the map d1 : E1 → E0 (5.2.1) of vector bundles on X ×X
constructed from L and show that if there is a closed embedding of X into Mθ(Q,J)
such that the tautological bundles on Mθ(Q,J) pull back to the line bundles in L on
X, then the cokernel of d1 is O∆ (Proposition 5.4). We finish the chapter by giving
conditions as to when X embeds into Mθ(Q,J) such that the tautological bundles
3
1.1. Structure of the Thesis
restrict to the line bundles; the conditions given depend on whether all of the line
bundles in L are nef or not.
Chapter 6
Chapter 6 explains how we compute the rest of the resolution
0→ Ek → · · · → E1d1→ E0
of O∆ and is motivated by the work of King [Kin97] on the smooth toric Fano sur-
faces. Setting A to be the endomorphism algebra End(⊕
i L−1i ) and T =
⊕i L
−1i , he
constructs the object T ∨L
⊠A T ∈ Db(X×X) from a minimal projective A,A-bimodule
resolution P • of A and shows that if T ∨L
⊠A T is quasi-isomorphic to O∆ in Db(X×X),
then L generates Db(X) (see Lemma 6.3). The final map in the minimal projective
A,A-bimodule resolution of A is determined by d1 and as King was working with 2-
dimensional varieties, P • could be calculated explicitly by only knowing the vertices
and arrows in the quiver for L (see Lemma 6.2); however, in general it is not known
how to compute P • for the algebra A. Our method utilises the idea of a toric cell
complex, introduced by Craw–Quintero-Velez [CQV12], to guess a minimal projective
A,A-bimodule resolution of A.
For the smooth toric Fano fourfolds in particular, by considering the pullback of L
to the total space of the canonical bundle Y := tot(ωX), the resulting rolled-up helix
algebra is expected to be a Calabi–Yau-5 (CY5) algebra for which we know the 0th,
1st and 2nd terms of its minimal projective bimodule resolution. The natural duality
inherent in a CY5 algebra then gives clues, via our guess as to what the corresponding
toric cell complex is, as to what the 3rd, 4th and 5th terms are. We sheafify the result,
restrict to X and then check that the resulting exact sequence of sheaves S• is indeed
a resolution of O∆ by using quiver moduli as explained in Chapter 5. Similarly for the
smooth toric Fano threefolds, we obtain an algebra expected to be CY4, in which case
the 0th term in its minimal projective bimodule resolution determines the 4th term, the
1st term determines the 3rd term and the 2nd term is self-dual. Using this method for
the collections of line bundles from the database contained in QuiversToricVarieties
[PN15a], we obtain resolutions of the diagonal for 88 of the smooth toric Fano fourfolds
and all 18 of the smooth toric Fano threefolds.
The final section in the chapter gives the framework for the second method we use
to show that a collection of line bundles on a smooth toric Fano fourfold is full, by
bringing together the concepts in Chapters 5 and 6.
Chapter 7
We present the main theorems of the thesis in Chapter 7. By considering the birational
geometry of the smooth toric Fano fourfolds (see Figure B.1) and choosing collections
L from a special set of line bundles on X0 as Uehara did for the toric Fano threefolds,
the pushforward of L onto a torus-invariant divisorial contraction X1 is automatically
full if L is full, and the pushforward coincides with the image of L under the map
γ : Pic(X0) → Pic(X1) (see Proposition 7.2 and Lemma 7.1). We can then check
4
1.1. Structure of the Thesis
that the collection on X1 is strong exceptional by ensuring that the necessary tensor
products of L avoid the preimage of the nnnvc-cones for X under the map γ, in addition
to the nnnvc-cones for X0; as outlined in Chapter 3, these preimages have a simple
description. Using this process, we obtain full strong exceptional collections on many of
the toric Fano fourfolds from the pushforward of collections on the birationally maximal
examples. With an additional computation, we then show that the tilting bundles we
obtain on the smooth toric Fano fourfolds as well as the tilting bundles Uehara exhibits
on the smooth toric Fano threefolds pull back to give tilting bundles that decompose as
a direct sum of line bundles on the total space of the canonical bundle for the variety.
Chapter 8
This chapter presents unanswered questions that have arisen from this thesis. For the 88
smooth toric Fano fourfolds and 18 smooth toric Fano threefolds for which we compute
a resolution of O∆ in Chapter 6, it is not known whether the toric cell complex exists
in any of these cases. Our calculations therefore lead us to the following conjecture:
Conjecture 8.1. Let X be a smooth toric Fano threefold or one of the 88 smooth toric
Fano fourfolds such that the given full strong exceptional collection L in the database
[PN15a] has a corresponding exact sequence of sheaves S• ∈ Db(X ×X). Let B denote
the rolled up helix algebra of A = End(⊕
L∈L L−1). Then the toric cell complex of B
exists and is supported on a real four or five-dimensional torus respectively. Moreover,
• the cellular resolution exists in the sense of [CQV12], thereby producing the min-
imal projective bimodule resolution of B;
• the object S• is quasi-isomorphic to T ∨L
⊠A T ∈ Db(X × X), where T :=
⊕L∈L L
−1 and T ∨L
⊠A T is the exterior tensor product over A of T ∨ and T .
The chapter also provides an example of a smooth toric Fano fourfold for which we
have failed to find a resolution of O∆ using the method in Chapter 6.
The Appendices
Appendix A consists of the article accompanying the Macaulay2 package QuiversToric-
Varieties [PN15a]. This package contains a database of the full strong exceptional col-
lections on n-dimensional smooth toric Fano varieties for 1 ≤ n ≤ 4, as well as many
of the computational tools used in the proofs of the theorems in the thesis. Appendix
B details how the the full strong exceptional collections on the smooth toric Fano va-
rieties of dimension ≤ 4 are obtained and has the divisorial contraction diagram for
the smooth toric Fano fourfolds, whilst Appendix C contains examples of smooth toric
Fano fourfolds for which we use the second method of generation to show that a given
strong exceptional collection of line bundles generates Db(X).
5
CHAPTER
TWO
BACKGROUND
This chapter provides the background material for the thesis and is divided into four
sections. The first section introduces toric geometry; given a lattice polytope, we
construct its corresponding fan Σ and show that this information can be used to create
a toric variety XΣ. Primitive collections and toric morphisms are defined and we
introduce the map γ : Pic(X0) → Pic(X1) induced from a torus-invariant divisorial
contraction X0 → X1.
In the second section we detail some basic properties of line bundles and recall
the classification of smooth toric Fano varieties in low dimensions. The third section
introduces the reader to the notions of a full strong exceptional collection and a tilting
object for the bounded derived category Db(X) of coherent sheaves on a smooth variety
X. We recall that the existence of a tilting object implies that Db(X) is equivalent to
the bounded derived category of a module category, and that tilting objects on varieties
Y and Z determine a tilting object on the product Y × Z.
The final section describes the construction of helices of sheaves on a variety and
gives Ginzburg’s definition of a Calabi-Yau algebra [Gin06].
2.1 Toric Geometry
For n ≥ 0, let M be a rank n lattice and define N := HomZ(M,Z) to be its dual lattice.
The realifications MR := M⊗ZR and NR := N⊗ZR are real vector spaces which contain
the underlying lattices and there exists a natural pairing 〈 , 〉 : MR × NR → R. The
convex hull of a finite set of lattice points in M defines a lattice polytope P ⊂MR and
its facets are the codimension 1 faces of P . We will assume that the dimension of P is
equal to the rank of M .
Definition 2.1. A convex polyhedral cone σ in NR is the set
{∑
u∈S
λuu | λu ≥ 0
}⊂ NR
for a given finite set S ⊂ NR. We say that σ is rational if additionally, S ⊂ N .
The theory of polytopes (see for example Cox–Little–Schenck [CLS11]) states that
every facet F in P has an inward-pointing normal nF that defines a one-dimensional
6
2.1. Toric Geometry
cone {λnF | λ ∈ R≥0} in NR. The cone is rational as P is a lattice polytope, so
it has a unique generator uF ∈ N . Given a ∈ R and a non-zero vector u ∈ NR we
have the affine hyperplane Hu,a := {m ∈ MR | 〈m,u〉 = a} and the closed half-space
H+u,a := {m ∈MR | 〈m,u〉 ≥ a}. As P is full-dimensional, each facet F defines a unique
number aF ∈ R such that F = HuF ,aF ∩ P and P ⊂ H+uF ,aF
. We can therefore use the
generators to completely describe P , using its unique facet presentation
P = {m ∈MR | 〈m,uF 〉 ≥ −aF for all facets F in P}.
Example 2.2. Let M = Z2 and P ⊂MR = R2 be the convex hull of the vertices
[0
1
],
[−1
1
],
[−1
−1
],
[2
−1
].
The polygon is shown in Figure 2.1, and its facet presentation is
P =
m ∈ R2
〈m, e1〉 ≥ −1
〈m, e2〉 ≥ −1
〈m,−e1 − e2〉 ≥ −1
〈m,−e2〉 ≥ −1
where {e1, e2} is the standard basis for R2.
b b b b
b b b b
b b b b
Figure 2.1: A lattice polytope in R2
If the origin of MR is an interior lattice point of P , then P has a dual polytope P ◦
which is defined to be the convex hull of the generators for the inward-pointing normal
rays of P :
P ◦ = Conv(uF | F is a facet of P ) ⊂ NR
The dual polytope determines a fan in NR:
Definition 2.3. Let F be a proper face of P ◦ with vertices {ui1 , . . . , uik}. The cone
σ(F ) is given by
σ(F ) := {λ1ui1 + . . .+ λkuik ∈ NR | λj ≥ 0, 1 ≤ j ≤ k}. (2.1.1)
The fan Σ(P ◦) ⊂ NR associated to P ◦ is given by the collection of cones
Σ = Σ(P ◦) := {0} ∪ {σ(F )}F(P ◦ (2.1.2)
where F runs over all proper faces of P ◦.
7
2.1. Toric Geometry
Let Σ(k) denote the set of k-dimensional cones in a fan Σ and we write τ � σ when
a cone τ is a face of a cone σ. The rays of Σ are the one-dimensional cones which, by
construction, are generated by the vectors uF for each facet F ⊂ P . We can use the
ray generators to define primitive collections and primitive relations which describe Σ
combinatorially.
Definition 2.4. A subset P = {ui1 , . . . , uik} of the set of ray generators V = {uF ∈
N | F is a facet of P} for Σ is a primitive collection if
(i) there does not exist a cone in Σ that contains every element of P and
(ii) any proper subset of P is contained in some cone of Σ.
The integral element s(P) = ui1 + . . . + uik is contained in some cone σ ∈ Σ with ray
generators {uj1 , . . . , ujm} and so can be uniquely written as a sum of the generators:
s(P) = c1uj1 + . . .+ cmujm , ci > 0, ci ∈ Z.
The linear relation
ui1 + . . .+ uik − (c1uj1 + . . .+ cmujm) = 0
between the ray generators of Σ is the primitive relation associated to the primitive
collection P.
Example 2.5. The dual polytope P ◦ to the lattice polytope in Example 2.2 has vertices
u1 =
[1
0
], u2 =
[0
1
], u3 =
[−1
−1
], u4 =
[0
−1
].
and is shown along with its corresponding fan Σ(P ◦) in Figure 2.5.
b b b
b b b
b b b
(a) The dualpolytope P
◦
b b b
b b b
b b b
u3u4
u2
u1
(b) The fan
Figure 2.2: The dual polytope in NR and its corresponding fan
The primitive relations for Σ(P ◦) are
• u1 + u3 − u4 = 0;
• u2 + u4 = 0.
Toric geometry (see e.g. Fulton [Ful93] or Cox–Little–Schenck [CLS11]) associates
to each fan Σ a toric variety XΣ such that M is the character lattice of the dense torus
8
2.1. Toric Geometry
T ∼= HomZ(M,C∗) in XΣ. For a cone σ ⊂ NR, its dual cone σ∨ is the set
σ∨ = {m ∈MR | 〈m,u〉 ≥ 0 for all u ∈ σ}.
Each cone σ ∈ Σ gives an affine variety
Uσ := Spec(C[σ∨ ∩M ]),
and for τ � σ we have Uτ ⊂ Uσ. The affine varieties Uσ for σ ∈ Σ are glued together
according to the arrangement of Σ, in which case we obtain the toric variety XΣ. If
the fan Σ is constructed from a polytope P ⊂MR as above, then we use XP to denote
the corresponding toric variety.
For a cone σ ∈ Σ define σ⊥ := {m ∈ MR | 〈m,u〉 = 0 for all u ∈ σ}. Each
cone σ determines a torus orbit O(σ) ⊂ XΣ and there is an isomorphism O(σ) ∼=
HomZ(σ⊥ ∩M,C∗) [CLS11, Lemma 3.2.5].
Proposition 2.6 (The Orbit-Cone Correspondence [CLS11]). Let XΣ be the toric
variety corresponding to a fan Σ ⊂ NR and n = dimNR. Then
1. There is a bijective correspondence
{cones in Σ} ←→ {T -orbits in XΣ}
σ ←→ O(σ)
2. For each cone σ ∈ Σ, dimO(σ) = n− dimσ.
3. The affine open subset Uσ is the union of the orbits
Uσ =⋃
τ�σ
O(τ).
4. τ � σ if and only if O(σ) ⊂ O(τ), and
O(τ) =⋃
τ�σ
O(σ).
where O(τ) is the closure of O(τ) in both the classical and Zariski topologies.
The Orbit-Cone Correspondence implies that for each ray ρ ∈ Σ(1), the closure of
the T -orbit O(ρ) is a torus-invariant divisor Dρ in XΣ. The lattice of torus-invariant
divisors in XΣ will therefore be denoted ZΣ(1) and the class group will be denoted
Cl(XΣ). We now have an exact sequence
0 −−−−→ M −−−−→ ZΣ(1) deg−−−−→ Cl(XΣ) −−−−→ 0 (2.1.3)
where the injective map is m 7→ Σρ∈Σ(1)〈m,uρ〉Dρ and the map deg sends the divisor
D to the isomorphism class of the rank one reflexive sheaf OXΣ(D). Henceforth, all of
the varieties that we consider in this thesis will be smooth, in which case every rank
one reflexive sheaf is invertible and so the class group Cl(XΣ) is isomorphic to the
9
2.1. Toric Geometry
Picard group Pic(XΣ). Note that XΣ is smooth if and only if for every cone σ ∈ Σ,
the minimal generators for σ form part of a Z-basis for N .
Example 2.7. The variety determined by the fan in Example 2.5 is the Hirzebruch
surface H1 = P(OP1 ⊕OP1(1)). From the fan, we see that H1 has four torus-invariant
points and four torus-invariant divisors. The exact sequence (2.1.3) for this variety is
0 −−−−→ Z2
1 0
0 1
−1 −1
0 −1
−−−−−−−−→ Z4
1 1 1 0
0 1 0 1
−−−−−−−−−−→ Z2 −−−−→ 0,
(2.1.4)
The Cox ring for XΣ is the semigroup ring SX := C[xρ | ρ ∈ Σ(1)] of NΣ(1) ⊂
ZΣ(1). The map deg induces a Cl(XΣ)-grading on SX , where the degree of a monomial∏ρ∈Σ(1) x
aρρ ∈ SX is the divisor class [
∑ρ∈Σ(1) aρDρ]. For α ∈ Cl(XΣ), we let (SX)α =
C[xρ | ρ ∈ Σ(1)]α denote the α-graded piece. Cox [Cox95, Proposition 3.1] defines
an exact functor from the category of Cl(XΣ)-graded SX-modules to the category of
quasi-coherent sheaves on XΣ:
{Cl(XΣ)-graded SX-modules} −→ Qcoh(XΣ) : M 7→ M. (2.1.5)
As XΣ is smooth, every coherent sheaf on XΣ is isomorphic to M for some finitely
generated Pic(XΣ)-graded SX-module M and two finitely generated Pic(XΣ)-graded
SX-modules determine isomorphic coherent sheaves if and only if they agree up to
saturation by the irrelevant ideal BX :=(∏
ρ�σ xρ | σ ∈ Σ)
[Cox95, Propositions
3.3, 3.5]. For α ∈ Pic(XΣ), we have the Pic(XΣ)-graded SX-module SX(α), where
(SX(α))β = (SX)α+β for β ∈ Pic(XΣ).
Morphisms between two toric varieties can be described by maps between their
associated fans that preserve the cone structure. For example, consider the blowup of
a torus-invariant subvariety. By the Orbit-Cone Correspondence, a k-codimensional
torus-invariant subvariety of a toric variety XΣ corresponds to a cone σ ∈ Σ(k), and
the blowup of this subvariety is the toric variety whose fan is the star subdivision of
σ. The star subdivision is a combinatorial process that introduces a new ray x with
generator uσ =∑
ρ∈σ(1) uρ and replaces Σ with
Σ∗σ,x := {τ ∈ Σ | σ � τ} ∪
⋃
σ�τ
Σ∗τ (σ) (2.1.6)
where Σ∗τ (σ) := {Cone(A) | A ⊆ {uσ} ∪ τ(1), σ(1) * A}. The map between fans
Σ∗σ,x → Σ determines the blowup
ϕ : X0 := XΣ∗σ,x−→ X1 := XΣ (2.1.7)
and induces a commutative diagram between the corresponding exact sequences (2.1.3)
10
2.2. Smooth Toric Fano Varieties
for the varieties:
0 −−−−→ M −−−−→ ZΣ∗
σ,x(1)degX0−−−−→ Pic(X0) −−−−→ 0
∥∥∥ β
y γ
y
0 −−−−→ M −−−−→ ZΣ(1)degX1−−−−→ Pic(X1) −−−−→ 0
(2.1.8)
where β projects away from the coordinate corresponding to the exceptional divisor
and γ is such that γ ◦ degX0= degX1
◦β.
Example 2.8. The Hirzebruch surface H1 is the blowup of P2 at a torus-invariant
point. The corresponding fans are given in Figure 2.3, where the cone with rays {u1, u3}
is star-subdivided and the exceptional divisor corresponds to ray u4. In this example,
the commutative diagram (2.1.8) is
0
0
Z2
Z2
Z3
Z4
Z
Z2
0
0
1 00 1−1 −10 −1
1 0 0 00 1 0 00 0 1 0
[1 1 1 00 1 0 1
]
1 00 1−1 −1
id
[1 1 1
]
[1 0
]
b b b
b b b
b b b
u3u4
u2
u1
(a) The fan forH1
b b b
b b b
b b b
u3
u2
u1
(b) The fanfor P2
Figure 2.3: The fans in the blowup H1 → P2
2.2 Smooth Toric Fano Varieties
From (2.1.3), a line bundle L on a smooth toric variety X is determined by some
torus-invariant divisor D =∑
ρ∈Σ(1) aρDρ.
11
2.2. Smooth Toric Fano Varieties
Definition 2.9. A divisor D =∑
ρ∈Σ(1) aρDρ is effective if aρ ≥ 0, for all ρ ∈ Σ(1).
Given an effective divisor D, we can consider whether L = OX(D) can be used
to embed X into a projective space. The space of global sections W := Γ(X,L) is
basepoint free if for every point p ∈ X, there is a section s ∈ W such that s(p) 6= 0.
The dual space W∨ determines the projective space P(W∨). Given a fixed p ∈ X and
a nonzero element vp in the fibre of p, there exists λs ∈ C for each section s ∈W such
that s(p) = λsvp. By defining the map lp ∈W∨ by lp(s) := λs, we obtain the morphism
φL : X → P(W∨)
p 7→ lp
Definition 2.10. The divisor D and line bundle L are very ample if D is basepoint
free and φL : X → P(W∨) is a closed embedding. If kD is very ample for some integer
k > 0, then D and L are ample.
Lemma 2.11. [CLS11, Theorem 6.1.15] On a smooth complete toric variety X, a line
bundle is ample if and only if it is very ample.
For an irreducible complete curve C on X and normalisation φ : C → C, the in-
tersection product D · C of D and C is defined as the degree of the line bundle
D · C := deg(φ∗OX(D)).
Definition 2.12. The divisor D is nef if D ·C ≥ 0 for every irreducible complete curve
C ⊆ X.
Every basepoint free divisor is nef, and when the fan of X has convex support of full
dimension, then a divisor is basepoint free if and only if it is nef [CLS11, Theorem
6.3.12]. The classes of nef divisors generate a cone Nef(X) in Pic(X)R.
Lemma 2.13. [CLS11, Theorem 6.3.22] Let X be a projective toric variety. Then the
divisor D is ample if and only if its class in Pic(X)R is in the interior of Nef(X).
On a variety X, the canonical bundle ωX is the line bundle that is the top exterior
power of the cotangent bundle. It is determined by the canonical divisor KX ; on a
toric variety, KX = −∑
ρ∈Σ(1)Dρ [CLS11, Theorem 8.2.3]. The dual to ωX is the
anticanonical bundle ω−1X and hence the anticanonical divisor is −KX =
∑ρ∈Σ(1)Dρ.
Definition 2.14. A smooth toric variety X whose anticanonical divisor −KX is ample
is called a smooth toric Fano variety.
A lattice polytope P in MR is reflexive if its facet presentation is
P = {m ∈MR | 〈m,uF 〉 ≥ −1 for all facets F in P}.
If P is reflexive then the origin of MR is the only interior lattice point of P and its
dual polytope is also a reflexive polytope. A polytope is smooth if its dual polytope
determines a smooth fan, and two reflexive polytopes P1, P2 ⊂MR are lattice equivalent
if P1 is the image of P2 under an invertible linear map of MR induced by an isomorphism
of M . Batyrev [Bat99] uses smooth reflexive polytopes to classify smooth toric Fano
varieties:
12
2.2. Smooth Toric Fano Varieties
Theorem 2.15. [Bat99, Theorem 2.2.4] If P is an n-dimensional smooth reflexive
polytope then XP is an n-dimensional smooth toric Fano variety. Conversely, If X is
an n-dimensional smooth toric Fano variety then there exists an n-dimensional smooth
reflexive polytope P such that XP∼= X. Moreover, if P1 and P2 are two smooth reflexive
polytopes then XP1∼= XP2 if and only if P1 and P2 are lattice equivalent.
Example 2.16. The lattice polygon in Example 2.2 is smooth and reflexive, hence
H1 is a smooth toric Fano surface. Choosing the basis {[D1], [D4]} for Pic(H1), the
effective cone Eff(H1) is generated by {[D1], [D4]} whilst the nef cone is generated by
{[D1], [D1 + D4]}. The class of the anticanonical bundle O(D1 + D2 + D3 + D4) ∼=
O(3D1 + 2D4) is contained in the interior of Nef(H1), so it is ample. The cones in
Pic(H1) are shown in Figure 2.4.
b b b b b
b b b b b
b b b b b
b b b b b
b b b b b
0
ωX
[D4]
[D1]
Nef(H1)
Eff(H1)
Figure 2.4: The effective cone and nef cone in Pic(H1)
We therefore refer to smooth reflexive lattice polytopes as Fano polytopes and as
Batyrev observed, there are finitely many Fano polytopes up to lattice equivalence
in each dimension [Bat82a]. There are five corresponding smooth toric Fano varieties
in dimension 2 that were known classically, whilst Watanabe-Watanabe [WW82] and
Batyrev [Bat82b] classified the 18 smooth toric Fano varieties in dimension 3. In di-
mension 4, Batyrev [Bat99] used primitive collections and relations to classify the Fano
polytopes and Sato [Sat00] completed the classification using toric blowups, bringing
the total number of 4-dimensional smooth toric Fano varieties to 124. Kreuzer and Nill
[KN09] calculated that there are 866 5-dimensional Fano polytopes up to lattice equiv-
alence, while Øbro [Øb07] presented an algorithm that has classified Fano polytopes in
dimensions up to 9.
Sato [Sat00] records the birational geometry between the smooth toric Fano four-
folds by computing toric divisorial contractions in terms of the primitive relations for
each variety. Figure B.1 in Appendix B is a diagram of the divisorial contractions be-
tween the smooth toric Fano fourfolds. There are 29 maximal toric Fano fourfolds with
regard to these divisorial contractions, and we call these varieties birationally maxi-
mal. A diagram showing the divisorial contractions between the smooth toric Fano
threefolds can be found in [Oda88, page 92] [WW82].
Remark 2.17. In [Sat00, Table 1], Sato states that there is a contraction from variety
K2 to variety H10. This contraction should be from variety K3 to H10.
13
2.3. Full Strong Exceptional Collections and Tilting Objects
2.3 Full Strong Exceptional Collections and Tilting Ob-
jects
For a set of objects S = {Si} in a triangulated category D, define 〈S〉 to be the smallest
triangulated subcategory of D containing S, closed under isomorphisms, taking cones
of morphisms and direct summands, and 〈S〉⊥ to be the full triangulated subcategory
of D containing objects F such that Hom(S,F) = 0 for all S ∈ S.
Definition 2.18. For a set of objects S = {Si} in D,
(i) S classically generates D if 〈S〉 = D,
(ii) S generates D if 〈S〉⊥ = 0.
Let Db(X) be the bounded derived category of coherent sheaves on a variety X. If X
is projective, Van den Bergh provides us with a set of objects that generate Db(X):
Lemma 2.19. [VdB04, Lemma 3.2.2] Let X be a projective variety of dimension n, L
an ample line bundle on X generated by global sections and a ∈ Z. If M in Db(X) is
such that Homi(La+j ,M) = 0 for all i and for 0 ≤ j ≤ n, then M = 0.
Definition 2.20. Let (E0, . . . , Er) be an ordered set of objects in Db(X).
(i) The set (E0, . . . , Er) is a strong exceptional collection if Hom(Ek, Ek) = C for all
k ∈ {0, . . . , r} and
Homi(Ek, Ej) = 0 when
{k > j, i = 0,
∀ k, j, i 6= 0.
(ii) A strong exceptional collection (E0, . . . , Er) in Db(X) is full if 〈E0, . . . , Er〉 =
Db(X).
Remark 2.21. The distinction between classical generation and generation becomes
irrelevant when using strong exceptional collections. To show that a strong exceptional
collection (E0, . . . , Er) is full, it is enough to show that 〈E0, . . . , Er〉⊥ = 0 as observed
by Bridgeland–Stern [BS10, Lemma C.1].
Definition 2.22. An object T in Db(X) is a tilting object if Homi(T ,T ) = 0 for i 6= 0
and 〈T 〉 = Db(X). If additionally T is a sheaf or vector bundle, then it is called a
tilting sheaf or tilting bundle respectively.
Given a full strong exceptional collection (E0, . . . , Er) of non-isomorphic objects in
Db(X), its sum⊕r
i=0 Ei is a tilting object.
For a tilting object T , let A = End(T ) and Db(A) be the bounded derived category
of finitely generated right A-modules. It was shown by Baer [Bae88] and Bondal [Bon90]
that in the case when X is a smooth projective variety, if the tilting object T exists
then we obtain an equivalence of categories
RHomX(T ,−) : Db(X) −→ Db(A). (2.3.1)
14
2.4. Helices and Calabi-Yau Algebras
Note that when T =⊕r
i=0 Ei is the direct sum of a full strong exceptional collection, the
Grothendieck group K0(X) of X is isomorphic to a rank r + 1 lattice; the equivalence
of derived categories above induces an isomorphism K0(X) ∼= K0(A), and the classes
of indecomposable projective A-modules corresponding to [Ei] for 0 ≤ i ≤ r freely
generate K0(A).
Example 2.23. King [Kin97] showed that the set of line bundles {O,O(D1),O(D1 +
D4),O(2D1 + D4)} is a full strong exceptional collection on H1. In Example 3.3 we
show that the collection is strong exceptional, whilst Examples 4.6 and 6.10 give two
different methods to prove that the collection is full.
For two smooth projective varieties Y and Z, let E ∈ Db(Y ) and F ∈ Db(Z). Define
EL
⊠ F := Lp∗1(E)L⊗ Lp∗2(F) ∈ Db(Y × Z)
where p1 and p2 are the natural projections of Y × Z onto its components. If we have
tilting objects on two varieties, then we immediately obtain a tilting object on the
product of the varieties:
Lemma 2.24. [Ueh14, Lemma 5.2] Let Y and Z be as above. If E ∈ Db(Y ) and
F ∈ Db(Z) are tilting objects, then EL
⊠ F is a tilting object for Db(Y × Z).
2.4 Helices and Calabi-Yau Algebras
An algebra B is homologically smooth if, viewed as a bimodule over itself, it has a
bounded resolution by finitely generated projective B,B-bimodules. We have the con-
travariant functor on the derived category of B,B-bimodules that maps objects:
M 7→M ! := RHomB,B-mod(M,B ⊗B).
Using the outer bimodule structure on B ⊗B when taking RHom results in M ! being
a B,B-bimodule using the inner structure. Any morphism f : M → N in the derived
category then induces a morphism f ! : N ! → M !. The following definition is due to
Ginzburg [Gin06]:
Definition 2.25. A homologically smooth algebra B is a Calabi-Yau algebra (CYd)
of dimension d if there exists a B,B-bimodule quasi-isomorphism
f : B∼=→ B![d] such that f = f ![d],
where [d] is the shift by d functor on the derived category of B,B-bimodules.
Proposition 2.26. [Gin06, Proposition 3.3.1] Let X be a smooth connected variety
which is projective over an affine variety and let E ∈ Db(X) be a tilting object. Then
the endomorphism algebra End(E) is a CYd algebra if and only if X is a Calabi-Yau
manifold of dimension d.
Following Bridgeland and Stern [BS10], we recall the definition of a geometric helix,
which can be used to give examples of CYd algebras.
15
2.4. Helices and Calabi-Yau Algebras
Definition 2.27. A sequence of coherent sheaves H = (Ei)i∈Z on a variety X is a helix
if
• for each i ∈ Z the thread (Ei+1, . . . , Ei+k) is a full exceptional collection,
• for each i ∈ Z, we have Ei−k = Ei ⊗ ωX .
If H satisfies the additional condition that for all s < t,
Homj(Es, Et) = 0 unless j = 0 (2.4.1)
then H is geometric. If H satisfies the weaker condition that each thread is a strong
exceptional collection, then H is said to be strong.
Remark 2.28. [BS10, Remark 3.2] If {E0, . . . , Ek−1} in a helix H is a full exceptional
collection, then each thread of H is a full exceptional collection.
The helix algebra A(H) associated to H is the graded algebra
A(H) :=⊕
t≥0
∏
j−i=t
Hom(Ei, Ej).
Twisting by ωX induces a Z-action
Hom(Ei, Ej)→ Hom(Ei−k, Ej−k)
and the subalgebra of invariant elements is known as the rolled-up helix algebra B(H).
Lemma 2.29. [BS10, Theorem 3.6] Let B = B(H) be the rolled-up helix algebra of a
geometric helix on an n-dimensional variety X and Y := tot(ωX) be the total space of
the canonical bundle. Then B is a graded CY(n + 1) algebra. Given a thread E ⊂ Hand the bundle map π : Y → X, there is an equivalence
ΦE : Db(B)→ Db(Y )
sending B to the object π∗(E), where E =⊕
Ej∈EEj . In particular, π∗(E) is a tilting
object for Y .
If X is a smooth toric Fano variety then the total space of its canonical bundle
Y = tot(ωX) is a smooth toric Calabi-Yau variety. The fan ΣY for Y can be constructed
from the fan ΣX for X; given a cone σ ∈ ΣX , define
σ := Cone((0, 1), (uρ, 1) | ρ ∈ σ(1)) ⊂ NR × R
Then the set of cones σ for σ ∈ ΣX and their faces form ΣY .
Example 2.30. The full strong exceptional collection on H1 given in Example 2.23
forms the geometric helix
H = {. . . ,O(−D1 −D4),O,O(D1),O(D1 +D4),O(2D1 +D4),O(3D1 + 2D4), . . .}
(see Example 7.8). By Lemma 2.29, the decomposable vector bundle π∗(O ⊕O(D1)⊕
O(D1 + D4) ⊕ O(2D1 + D4)) is therefore a tilting bundle on tot(ωH1). The slice at
height 1 of the fan for tot(ωH1) is given in Figure 2.5.
16
2.4. Helices and Calabi-Yau Algebras
b b b
b b b
b b b
Figure 2.5: The slice at height 1 of the fan for tot(ωH1)
17
CHAPTER
THREE
STRONG EXCEPTIONAL COLLECTIONS ON SMOOTH
TORIC VARIETIES
The combinatorics of the fan Σ for a toric variety XΣ allow us to computationally
determine whether a collection of line bundles on XΣ is strong exceptional. The first
section in this chapter explains the construction of the not-necessarily non-vanishing
cohomology cones (nnnvc-cones) in the Picard lattice for XΣ, described by Eisenbud,
Mustata and Stillman [EMS00], which we utilise to achieve this goal.
The second section considers how the nnnvc-cones behave under torus-invariant
divisorial contractions and the effect these contractions have on strong exceptional
collections. We find that for a chain of contractions X0 → X1 → · · · → Xt, the
preimage in Pic(X0) of the nnnvc-cones for {X1, . . . ,Xt} under the induced maps
between the Picard lattices have a simple description in terms of the nnnvc-cones for
X0 – see Proposition 3.13. Proposition 3.14 then shows that this simplicity has a
consequence for the image of strong exceptional collections on X0 under the Picard
lattice maps when the varieties considered are smooth toric Fano fourfolds.
3.1 nnnvc-Cones
To check that a collection of effective line bundles {L0 := OX , L1, . . . , Lr} on a smooth
toric variety X is strong exceptional, one needs to check that H i(X,L−1s ⊗ Lt) ∼=
Homi(Ls, Lt) = 0 for i > 0 and 0 ≤ s, t ≤ r. Eisenbud, Mustata and Stillman
[EMS00] introduced a method to determine when the cohomology of a line bundle on X
vanishes by considering whether the line bundle avoids certain affine cones constructed
in Pic(X)R. We recall the construction of these cones below.
Let X be an n-dimensional toric variety with fan Σ, |Σ| be the support of the fan in
NR and recall that Σ(1) denotes the set of rays in Σ. For I ⊆ Σ(1), let YI be the union
of the cones in Σ having all edges in the complement of I. Using reduced cohomology
with coefficients in C we have
H iYI
(|Σ|) := H i(|Σ|, |Σ|\YI) = H i−1(|Σ|\YI), (3.1.1)
where the last equality holds for i > 0 as |Σ| is contractible.
18
3.1. nnnvc-Cones
An element of
HΣ := {I ⊆ Σ(1) | H iYI
(|Σ|) 6= 0 for some i > 0} (3.1.2)
is called a forbidden set. Define
pI ∈ ZΣ(1), where (pI)ρ =
{−1 if ρ ∈ I
0 if ρ /∈ I(3.1.3)
and
CI ={x = (xρ) ∈ ZΣ(1) | xρ ≤ 0 if ρ ∈ I, xρ ≥ 0 if ρ /∈ I
}. (3.1.4)
Setting LI := CI +pI ⊆ ZΣ(1) we see that LI ⊂ CI and LI = {x ∈ ZΣ(1) | neg(x) = I},
where neg(x) = {ρ ∈ Σ(1) | xρ < 0} ⊆ Σ(1).
Eisenbud, Mustata and Stillman show that for i ≥ 1, the cohomology of all twists
of the structure sheaf
H i∗(OX) :=
⊕
α∈Pic(X)
H i(X,OX (α))
is isomorphic as a graded SX-module to the local cohomology H iBX
(SX) of the Cox
ring [EMS00, Proposition 2.3(a)]. The ring SX has a finer grading by ZΣ(1) that is
compatible with the Pic(X)-grading, and this descends to give a grading on H iBX
(SX).
For any x,y ∈ ZΣ(1) such that neg(x) = neg(y), we have H i∗(OX)x ∼= H i
∗(OX)y[EMS00, Theorem 2.4].
Lemma 3.1. [EMS00, Theorem 2.7] Let x ∈ ZΣ(1) and I = neg(x). Then
H i∗(OX)x ∼= H i
YI(|Σ|).
Now, if D =∑
ρ∈Σ(1) xρDρ is the toric divisor that corresponds to x ∈ ZΣ(1), then
H i∗(OX)x ∼= H i(X,OX (D)). It therefore follows that x lies in LI for some I ∈ HΣ if
and only if
H i(X,OX (D)) 6= 0, for some i > 0. (3.1.5)
The convex hull of the set of lattice points LI forms an affine cone in RΣ(1).
Definition 3.2. Let I ∈ HΣ and consider the cone in RΣ(1) determined by the convex
hull of LI . The image in Pic(X)R of this cone under the map deg is a not-necessarily
non-vanishing cohomology cone (nnnvc-cone) and is denoted by ΛI .
We say that ΛI is a not-necessarily non-vanishing cohomology cone as the semigroup
corresponding to the image of LI under the map deg may not be saturated. In par-
ticular, if α ∈ ΛI then it is not necessarily the case that H i(X,OX (α)) 6= 0 for some
i > 0, but if α is not in ΛI for any I ∈ HΣ then H i(X,OX (α)) = 0 for all i > 0. Given
a collection of line bundles {L0, L1, . . . , Lr} on X, it follows that if L−1s ⊗Lt avoids all
of the nnnvc-cones for all 0 ≤ s, t ≤ r, then the collection is strong exceptional.
Example 3.3. Using the fan for H1 in Example 2.5, we see that the forbidden sets
19
3.2. Cones Affected by Blow Ups
are {{1, 3}, {2, 4}, {1, 2, 3, 4}}. The corresponding sets of lattice points in ZΣ(1) are
L{1,3} = {x ∈ ZΣ(1) | x1, x3 ≤ −1, x2, x4 ≥ 0}
L{2,4} = {x ∈ ZΣ(1) | x2, x4 ≤ −1, x1, x3 ≥ 0}
L{1,2,3,4} = {x ∈ ZΣ(1) | x1, x2, x3, x4 ≤ −1}.
The nnnvc-cones are given by the equations
Λ{1,3} = {a ∈ Pic(H1)R | a1 − a2 ≤ −2, −a2 ≤ 0}
Λ{2,4} = {a ∈ Pic(H1)R | −a1 + a2 ≤ −1, a2 ≤ −2}
Λ{1,2,3,4} = {a ∈ Pic(H1)R | a1 ≤ −3, a2 ≤ −2}.
Let L0 = O, L1 = O(D1), L2 = O(D1 + D4) and L3 = O(2D1 + D4). Then each
line bundle Li ⊗ L−1j , 0 ≤ i, j ≤ 3 is denoted by a ‘×’ in Figure 3.1, which also
displays the nnnvc-cones. As the line bundles avoid the nnnvc-cones, the collection
L = {L0, . . . , L3} is strong exceptional.
b b b b b b b
b b b b b b b
b b b b b b b
b b × × × b b
b b b × × × b
b b b b × × ×
b b b b b b b
0
Λ{1,2,3,4} Λ{2,4}
Λ{1,3}
(-3,-2) (-1,-2)
(-2,0)
Figure 3.1: The nnnvc-cones in Pic(H1)
3.2 Cones Affected by Blow Ups
Assume that we have a chain of torus-invariant divisorial contractions X := X0 →
X1 → · · · → Xt between smooth toric varieties and let L = {L0, L1, . . . , Lr} be a
collection of non-isomorphic line bundles on X with corresponding vectors {v0, . . . , vr}
in Pic(X)R. By (2.1.8) we have maps between the Picard lattices
Pic(X0)γ1−→ Pic(X1)
γ2−→ · · ·
γt−→ Pic(Xt). (3.2.1)
For ease of notation we set:
20
3.2. Cones Affected by Blow Ups
• γ(i→j) to be the composition of maps γj ◦ γj−1 ◦ · · · ◦ γi+1 for 0 ≤ i < j ≤ t;
• LXkto be the set of non-isomorphic line bundles on Xk in the image of γ(0→k)(L),
for 1 ≤ k ≤ t;
• ΛI,Xkto be the preimage in Pic(X0)R of the nnnvc-cone ΛI for Xk under the map
γ(0→k), for 1 ≤ k ≤ t;
• Ck ⊂ Pic(X0)R to be the preimage of all nnnvc-cones for Xk under the map
γ(0→k) for 1 ≤ k ≤ t, and C0 ⊂ Pic(X0)R to be the nnnvc-cones for X0.
By the construction of the sets Ck, we have the following result:
Lemma 3.4. If
vi − vj /∈t⋃
k=0
Ck
for all 0 ≤ i, j ≤ r then L is strong exceptional on X and LXkis strong exceptional on
Xk, for 1 ≤ k ≤ t.
It will be shown in this section that the preimage of the nnnvc-cones for Xk under
γ(0→k) is closely related to the nnnvc-cones for X0.
Lemma 3.5 (Forbidden sets duality). Let I ( Σ(1) and set I∨ = Σ(1)\I. If I ∈ HΣ,
then I∨ ∈ HΣ.
Proof. It is enough to show that the line bundle O(−∑
ρ∈I∨ Dρ) corresponding to pI∨
has non-vanishing higher cohomology. Let D := −∑
ρ∈I Dρ be the torus-invariant Weil
divisor corresponding to pI . By assumption, H i(X,O(D)) 6= 0 for some 0 < i < n. By
Serre duality and the fact that the canonical divisor is KX = −∑
ρ∈Σ(1)Dρ,
0 6= H i(X,O(D))∨ ∼= Hn−i (X,O(KX −D)) = Hj(X,O(∑
ρ∈Σ(1)
bρDρ)) (3.2.2)
where bρ = −1− aρ. But
aρ =
{−1 if ρ ∈ I
0 if ρ /∈ I⇒ bρ =
{0 if ρ ∈ I
−1 if ρ /∈ I.(3.2.3)
Therefore, (bρ) = pI∨ and as Hj(X,O(∑
ρ∈Σ(1) bρDρ)) 6= 0 for some 0 < j < n, we
have I∨ ∈ HΣ.
Remark 3.6. In the following lemmas, we change convention by setting YI to be the
union of the cones in Σ having all edges in I ( Σ(1). Due to the duality statement of
Lemma 3.5, this does not affect the outcome of Proposition 3.13.
Continuing with the notation in (2.1.7), it is clear that for
Cσ :=⋃
σ�τ∈Σ
τ (3.2.4)
21
3.2. Cones Affected by Blow Ups
we have
Σ\Cσ = Σ∗\⋃
σ�τ
Σ∗τ (σ) (3.2.5)
and so we only need to consider Cσ when determining how the cones of Σ change after
the blow up of σ ∈ Σ.
Lemma 3.7. Let ∅ 6= I ⊆ Σ(1)\Cσ(1). Then I ∪ {x} ∈ HΣ∗.
Proof. Firstly, assume for some I ⊂ Σ(1) that there exists a ray τ ( YI such that
τ ∩ σ = {0} for all cones τ 6= σ ⊂ YI . By considering YI ⊂ |Σ| ∼= Rn for n > 2, we can
construct a loop around τ that is not contractible in |Σ|\YI ; if n = 2, then |Σ|\YI is a
disconnected space. Thus I ∈ HΣ.
Now assume ∅ 6= I ⊆ Σ(1)\Cσ(1). By the construction of Σ∗ we have x ∩ σ = {0}
for any cone x 6= σ ⊂ Y Σ∗
I∪{x} and x 6= Y Σ∗
I∪{x}, so I ∪ {x} ∈ HΣ∗ by the observation
above.
By Lemma 3.5, (I ∪ {x})∨ ∈ HΣ∗ for ∅ 6= I ⊆ Σ(1)\Cσ(1). But (I ∪{x})∨ = J ∪Cσ(1)
for some J ( Σ(1)\Cσ(1), so we have the corollary:
Corollary 3.8. If I ( Σ(1)\Cσ(1), then I ∪ Cσ(1) ∈ HΣ∗.
Lemma 3.9. If I ∈ HΣ and I ∩ Cσ(1) = ∅ then I, I ∪ {x} ∈ HΣ∗.
Proof. Let I ∈ HΣ such that I ∩ Cσ(1) = ∅. By Lemma 3.7, I ∪ {x} ∈ HΣ∗. As
I ∩ Cσ(1) = ∅, then Y Σ∗
I = Y ΣI and |Σ∗| = |Σ|, so H i
Y Σ∗
I
(|Σ∗|) = H iY ΣI
(|Σ|) for all i.
Thus I ∈ HΣ∗ .
Again by duality, we have the corollary:
Corollary 3.10. If I ∈ HΣ is such that Cσ(1) ⊆ I, then I, I ∪ {x} ∈ HΣ∗.
Lemma 3.11. If I ∈ HΣ then either I ∈ HΣ∗ or I ∪ {x} ∈ HΣ∗.
Proof. We have shown that the statement holds if I∩Cσ(1) = ∅ and dually if Cσ(1) ⊆ I.
Therefore, assume that I ∩Cσ(1) 6= ∅ and Cσ(1) * I. There are two cases to consider:
Case 1: (σ(1) * I). Any subset S ⊆ τ(1) of any cone τ ⊂ Σ forms a cone in Σ as Σ is a
smooth fan. From this and the fact that σ(1), {x} * I we see that Y ΣI = Y Σ∗
I by
the construction of Σ∗. Therefore I ∈ HΣ ⇒ I ∈ HΣ∗.
Case 2: (σ(1) ⊆ I). By duality I∨ ∈ HΣ and I∨ ∩ σ(1) = ∅, so I∨ ∈ HΣ∗ by Case 1.
Applying duality again we have I ∪ {x} = (I∨)∨ ∈ HΣ∗.
Remark 3.12. It is not always the case that I ∈ HΣ ⇒ I, I ∪ {x} ∈ HΣ∗ (see Example
3.17).
Recalling the chain of linear maps (3.2.1), we have a simple description of the preimage
in Pic(X)R of the nnnvc-cones for the variety Xt using the nnnvc-cones for X. Let
{E1, . . . , Et} be the exceptional divisors from the blow ups in (3.2.1). The list can be
extended to give a basis {[E1], . . . , [Et], y1, . . . , ys} of Pic(X)R.
22
3.2. Cones Affected by Blow Ups
Proposition 3.13. Let ΛI ⊆ Pic(Xt)R be a nnnvc-cone for Xt in (3.2.1). There
exists a nnnvc-cone ΛI′ ⊆ Pic(X)R for X with the following property: describe ΛI′
by the intersection of closed half-spaces in Pic(X)R given by equations ai1[E1] + . . . +
ait[Et] + ait+1y1 + . . . + ait+sys ≤ ai where ai1, . . . , ait+s, a
i ∈ R are fixed and i is in an
indexing set S. Then the preimage ΛI,Xt is the intersection of the closed half spaces
ait+1y1 + . . .+ ait+sys ≤ ai, i ∈ S.
Proof. We first show the statement for the blowup ϕ : XΣ∗σ,x−→ XΣ from (2.1.7). Let
ΛI be a nnnvc-cone for XΣ determined by the forbidden set I ⊂ Σ(1). By Lemma 3.11
there exists a nnnvc-cone ΛI′ ⊆ Pic(XΣ∗σ,x
)R for XΣ∗σ,x
such that its defining forbidden
set I ′ is either I ∪ {x} ⊂ Σ∗σ,x(1) or I ⊂ Σ∗
σ,x(1). By construction of LI ⊆ ZΣ(1)
and LI′ ⊆ ZΣ∗
σ,x(1), the closed half spaces in Pic(XΣ∗σ,x
)R describing ΛI′ are given by
equations ai0[E] + ai1y1 + . . . + aisys ≤ ai for fixed ai0, . . . , ais, a
i ∈ R and i ∈ S, whilst
those in Pic(XΣ)R describing ΛI are ai1y1 + . . .+ aisys ≤ ai. The map β in (2.1.8) is a
projection away from the coordinate corresponding to the exceptional divisor E, hence
the map γ is a projection away from the exceptional divisor class [E] as (2.1.8) is a
commutative digram. Therefore, the preimage ΛI,XΣis given by the the intersection
of halfspaces with equations ai1y1 + . . .+ aisys ≤ ai. By repeated application of Lemma
3.11, we obtain the required result for a chain of blowups (3.2.1).
The simplicity of the preimage of nnnvc-cones under blowups can help explain why
the following proposition holds. Recall that a smooth toric Fano variety X is called
birationally maximal if there does not exist a smooth toric Fano variety X ′ with blowup
X ′ → X.
Proposition 3.14. Let X be a birationally maximal smooth toric Fano fourfold and
r + 1 = rank(K0(X)). There exists a strong exceptional collection of line bundles L =
{L0, . . . , Lr} on X such that for every chain of torus-invariant divisorial contractions
X → X1 → · · · → Xt from Figure B.1, the set of line bundles LXion Xi is strong
exceptional, for 1 ≤ i ≤ t. A database of these collections can be found in [PNb].
Proof. Given a birationally maximal smooth toric Fano fourfold X and a chain of
divisorial contractions between {X0 := X,X1, . . . ,Xt}, we construct the preimage Ci
in Pic(X)R of the nnnvc-cones for each contraction Xi using the QuiversToricVarieties
package [PN15a]. A computer search then finds line bundles {L0, L1, . . . , Lr} on X
with corresponding vectors {v0, . . . , vr} in Pic(X)R such that vj − vk avoids Ci for all
0 ≤ j, k ≤ r and 0 ≤ i ≤ t.
Remarks 3.15.
(i) The collections given in Proposition 3.14 are not necessarily the same collections
given by Theorem 7.4. In particular, not all of them have been shown to be full.
(ii) If two toric varieties X1 and X2 have the same primitive collections, then they have
the same forbidden sets up to a suitable ordering of the rays of ΣX1 and ΣX2 . It is
therefore often the case that given a suitable basis of Pic(X1)R and Pic(X2)R, if the line
bundles corresponding to a list of integral points {vj}j∈J ⊂ Rd ∼= Pic(X1)R is strong
exceptional on X1, then the collection of line bundles corresponding to the same list
{vj}j∈J ⊂ Rd ∼= Pic(X2)R is strong exceptional on X2.
23
3.2. Cones Affected by Blow Ups
Example 3.16. Example 2.8 describes the fans for the varieties in the blowup H1 → P2
and the map γ. The only forbidden set for P2 is {1, 2, 3}, determining the nnnvc-cone
Λ{1,2,3} = {a ∈ Pic(P2)R | a1 ≤ −3}.
The preimage under γ of this cone in Pic(H1) is
Λ{1,2,3},P2 = {a ∈ Pic(H1)R | a1 ≤ −3}
and is pictured in Figure 3.2 along with the nnnvc-cones for H1. In particular,
Λ{1,2,3},P2 is a supporting half-space of Λ{1,2,3,4}. Using the same collection of line
bundles L as in Example 3.3, Figure 3.2 shows each Li⊗L−1j , 0 ≤ i, j ≤ 3 as a ‘×’. We
see that these line bundles avoid Λ{1,2,3},P2 as well as the nnnvc-cones for H1, hence L
is strong exceptional on H1 and LP2 is strong exceptional on P2.
b b b b b b b
b b b b b b b
b b b b b b b
b b × × × b b
b b b × × × b
b b b b × × ×
b b b b b b b
0
Λ{1,2,3,4} Λ{2,4}
Λ{1,3}
Λ{1,2,3,4},P2
(-1,-2)
(-2,0)
Figure 3.2: The nnnvc-cones for the blowup H1 → P2
Example 3.17. The smooth toric Fano fourfold X0 := E1 has ray generators
u0 =
1
0
0
0
, u1 =
0
1
0
0
, u2 =
0
0
1
0
, u3 =
0
0
0
1
, u4 =
−1
0
0
0
, u5 =
3
−1
−1
−1
, u6 =
2
−1
−1
−1
for its fan ΣX0 . The blowup φ : E1 → B1 of the smooth toric Fano fourfold B1 (see
Figure B.1) has the exceptional divisor E = D6 labelled by the ray generator u6.
Note that X1 := B1 has the fan ΣX1 with ray generators {u0, . . . , u5}. We take the
corresponding divisor classes {[D0], [D1], [E]} to be a basis for Pic(X0), and the linear
equivalences between the divisors for X0 are D1 ∼ D2 ∼ D3, D4 ∼ D0+3D1−E, D5 ∼
D1 − E. The linear equivalences between the divisors for X1 are D′1 ∼ D′
2 ∼ D′3 ∼
D′5, D
′4 ∼ D
′0 + 3D′
1. The forbidden sets for X0 are
24
3.2. Cones Affected by Blow Ups
nnnv i-th Cohomology Cones Forbidden Sets
1 {0, 4}, {4, 5}, {0, 4, 5},
{0, 6}, {0, 4, 6}
2
3 {1, 2, 3, 5}, {1, 2, 3, 4, 5},
{1, 2, 3, 6}, {0, 1, 2, 3, 6},
{1, 2, 3, 5, 6}
4 {0, 1, 2, 3, 4, 5, 6}
and the forbidden sets for X1 are
nnnv i-th Cohomology Cones Forbidden Sets
1 {0, 4}
2
3 {1, 2, 3, 5}
4 {0, 1, 2, 3, 4, 5}
In this example we see that for the forbidden set I ∈ {{0, 4}, {1, 2, 3, 5}} for X1, both
I and I ∪{6} are forbidden sets for X0, whilst for the forbidden set I = {0, 1, 2, 3, 4, 5}
for X1, only I ∪ {6} is a forbidden set for X0. Now
Λ{0,4},X1∩ Pic(X0) = (Λ{0,4} ∪ Λ{0,4,6}) ∩ Pic(X0)
and
Λ{1,2,3,5},X1∩ Pic(X0) = (Λ{1,2,3,5} ∪ Λ{1,2,3,5,6}) ∩ Pic(X0).
Thus for a strong exceptional collection of line bundles L on X0, only
Λ{0,1,2,3,4,5},X1
provides a restriction for the distinct line bundles in the image of γ(L) to be strong
exceptional on X1. The cone Λ{0,1,2,3,4,5,6} is given by the system of equations
a1 ≤ −2
a2 ≤ −7
a2 + a3 ≤ −6
,
a1a2a3
∈ Pic(X0)R
in Pic(X0)R, whilst Λ{0,1,2,3,4,5},X1is given by the system of equations
{a1 ≤ −2
a2 ≤ −7,
a1a2a3
∈ Pic(X0)R
as expected by Proposition 3.13.
25
CHAPTER
FOUR
GENERATION OF Db(X) : THE FROBENIUS MORPHISM
(METHOD 1)
Let X be an n-dimensional smooth toric variety and L a strong exceptional collection
on X. In this thesis we present two different methods to show that L is full. The first
method depends on the Frobenius morphism and follows Uehara’s approach [Ueh14] to
generation of the derived category by line bundles on the smooth toric Fano threefolds.
The first section in this chapter recalls the construction of the Frobenius morphism and
how the Frobenius pushforward of a line bundle on a toric variety can be computed.
The second section introduces the framework we use to show that a given collection of
line bundles is full using the Frobenius morphism; we call this approach Method 1.
4.1 The Frobenius Morphism
Fix a positive integer m and let N ′ be the lattice N ′ := 1mN with dual M ′. The
Frobenius morphism is the finite surjective toric morphism Fm : X → X induced from
the natural inclusion fm : N → N ′ which maps a cone in NR to one in N ′R with the
same support. Thomsen [Tho00] shows that in characteristic p > 0, the Frobenius
pushforward (Fm)∗(L) of a line bundle L on X splits into a finite direct sum of line
bundles. He provides an algorithm to compute these line bundles, which we explain
below in the characteristic 0 setting by following [LM11] and [Ueh14].
Let Σ be the fan for X and set d := |Σ(1)|. From (2.1.3), a vector w ∈ ZΣ(1)
determines the line bundle L = OX(∑wiDi). To compute (Fm)∗(L), fix a maximal
cone σ ∈ Σ and set
P pm := {v ∈ Zp | 0 ≤ vi < m}. (4.1.1)
Define A := (uρ)ρ∈Σ(1) ∈M(d, n) to be the matrix whose rows are the ray generators uρin Σ. As σ is maximal and Σ is smooth, the corresponding matrix Aσ := (uρ)ρ∈σ(1) ∈
M(n, n) is invertible. Define the restriction w to σ as wσ := (wρ)ρ∈σ(1) ∈ Zn. For
v ∈ Pnm, the vectors qm(v,w, σ) ∈ ZΣ(1) and rm(v,w, σ) ∈ P d
m are uniquely determined
by the equation
AA−1σ (v−wσ) + w = mqm(v,w, σ) + rm(v,w, σ). (4.1.2)
26
4.1. The Frobenius Morphism
Note that if we set
x :=AA−1
σ (v−wσ) + w
m,
then the vector qm(v,w, σ) is given by ⌊x⌋; that is, the vector whose entries ⌊xρ⌋ ∈ Zare given by the round-down xρ − 1 < ⌊xρ⌋ ≤ xρ. Finally, define the Weil divisor
Dmv,w,σ :=
∑ρ∈Σ(1) q
mρ (v,w, σ)Dρ.
Lemma 4.1. [LM11, Proposition 3.1] The Frobenius push-forward of L = OX(∑wρDρ)
is
(Fm)∗(L) =⊕
v∈Pnm
OX(Dmv,w,σ). (4.1.3)
Proof. Fix an n-dimensional cone σ ∈ Σ(n). Set C[Sσ] := C[xρ | ρ ∈ σ(1)] and consider
the torus-invariant open affine subvariety Uσ = SpecC[Sσ]. The map Fm induces a
map of C-algebras
F#m : C[Sσ]→ C[Sσ], xu 7→ xmu.
We have the M -graded decomposition
C[Sσ] ∼=⊕
m∈M
C[Sσ]m
and the embedding from F#m extends this grading to a finer M -grading of C[Sσ].
Shifting C[Sσ] by a degree wσ gives a rank one M -graded C[Sσ]-module and with
respect to the refined grading we have the decomposition via F#m
C[Sσ](wσ) ∼=⊕
v∈Pnm
C[Sσ](v −wσ),
as Pnm gives a set of representatives of classes in M/mM . Now, in order to obtain
a standard M -grading, we use the round-down to choose suitable representatives, in
which case
C[Sσ](v −wσ) ∼= C[Sσ]
(⌊AA−1
σ (v−wσ) + w
m
⌋)∼=
⊕
u∈M
C[Sσ]u+
⌊
AA−1σ (v−wσ)+w
m
⌋.
This isomorphism can be globalised as follows. For every maximal cone σ ∈ Σ(n), there
exists wσ such thatOX(D) is represented by theM -graded C[Sσ]-module Γ(Uσ,OX(D)) ∼=
C[Sσ](wσ). As before, we have
C[Sσ](wσ) ∼=⊕
v∈Pnm
C[Sσ]
(⌊AA−1
σ (v−wσ) + w
m
⌋).
Division by m and rounding down lead to compatible M -graded decompositions on
each affine piece. Now, the pushforward (Fm)∗(OX(D)) is locally free and
(Fm)∗(OX (D)) ∼=⊕
χ∈M/mM
Oχ
where the Oχ are invertible sheaves. As⌊AA−1
σ (v−wσ)+w
m
⌋is independent of the repre-
27
4.2. Method 1
sentative for the class in M/mM , we can choose one fixed representative v ∈ Pnm for
each χ ∈M/mM . Hence Oχ∼= OX(Dm
v,w,σ) and the result follows.
Following Thomsen [Tho00], Uehara notes that (Fm)∗(L) does not depend on the choice
of the maximal cone σ [Ueh14, Lemma 3.4]. We can assume the primitive ray generators
of σ form the standard basis of Zn, in which case
qmρ (v,w) = ⌊uρ(v−wσ) + wρ
m⌋. (4.1.4)
Set
D(OX(D))m := {L ∈ Pic(X) | L is a direct summand of (Fm)∗(OX(D))}. (4.1.5)
Uehara [Ueh14, Lemma 3.5] also shows that the set
D(OX(D)) :=⋃
m>0
D(OX(D))m (4.1.6)
is finite. For brevity, we denote Dm := D(OX)m, the set of line bundles in (Fm)∗(OX).
Note that we can use Dm to find strong exceptional collections of line bundles on X:
Lemma 4.2. [Ueh14, Lemma 3.8(i)] For any fixed positive integer m, the set of line
bundles {L ∈ Dm | L−1 is nef} ⊆ Dm is a strong exceptional collection on X.
Example 4.3. Set m = 10 and let v = (x, y) ∈ P 2m. Using the rays of the fan given in
Example 2.5 for H1, the solutions to
qm(v,0) =
⌊ xm⌋
⌊ ym⌋
⌊−x−ym ⌋
⌊−ym ⌋
=
0
0
⌊−x−ym ⌋
⌊−ym ⌋
determine the line bundles in the Frobenius pushforward of the structure sheaf for H1.
We find that
Dm = {O, O(−D3), O(−D3 −D4), O(−2D3 −D4)}.
The inverse of every line bundle in this collection is nef, so Dm is strong exceptional
by Lemma 4.2. Note that this collection is dual to the collection given in Example 3.3.
4.2 Method 1
We can use the Frobenius morphism to find sets of line bundles that generate Db(X).
Lemma 4.4. [Ueh14, Lemma 5.1] Let f : X → Y be a proper morphism between
smooth varieties. Assume that E generates Db(X) and OY is a direct summand of
Rf∗OX . Then Rf∗E generates Db(Y ).
28
4.2. Method 1
Proposition 4.5. Let X be a smooth toric Fano variety of dimension n and L be a
strong exceptional collection of line bundles on X. If the set of line bundles
Dgenm :=
⋃
0≤i≤n
D(ω−iX )m (4.2.1)
is contained in 〈L〉 for some positive integer m, then L is full.
Proof. As X is Fano, the anticanonical bundle ω−1X is ample and so Lemma 2.19 implies
that⊕n
i=0 ω−iX is a generator for Db(X). The Frobenius morphism Fm is proper so⋃
0≤i≤nD(ω−i)m generates Db(X) by Lemma 4.4. By Remark 2.21, it follows that L
classically generates Db(X) and hence is full.
To show that Dgenm ⊂ 〈L〉 for some m > 0, we use exact sequences of line bundles to
generate objects in 〈L〉; for examples of these calculations on the toric Fano threefolds
see [Ueh14] or [BT09]. This process is easier when the line bundles in Dgenm are close
together in Pic(X), which occurs when the value of m is large. However, the larger the
value of m, the longer it takes to compute Dgenm , so in practice m is often chosen by
trial and error.
Example 4.6. Fix m = 10 and let v = (x, y) ∈ P 2m. Recall that the anticanonical
divisor −KX for H1 corresponds to w = (1, 1, 1, 1) ∈ ZΣ(1). The set Dm for H1 was
calculated in Example 4.3; to find D(ω−1)m and D(ω−2)m, we calculate
qm(v,w) =
⌊ (x−1)+1m ⌋
⌊ (y−1)+1m ⌋
⌊ (−x−y)+1m ⌋
⌊ (−y)+1m ⌋
=
0
0
⌊−x−y+1m ⌋
⌊−y+1m ⌋
and
qm(v, 2w) =
⌊ (x−2)+2m ⌋
⌊ (y−2)+2m ⌋
⌊ (−x−y)+2m ⌋
⌊ (−y)+2m ⌋
=
0
0
⌊−x−y+2m ⌋
⌊−y+2m ⌋
respectively. The result is D(ω−2)m = D(ω−1)m = Dm and Example 4.3 shows that
Dm = {O, O(−D3), O(−D3−D4), O(−2D3−D4)} is a strong exceptional collection.
As Dgenm = Dm, then Dm is a full strong exceptional collection for H1 by Proposition
4.5.
Example 4.7. We use Method 1 to show that a given collection of line bundlesgenerates Db(X) when X is the smooth toric Fano fourfold I1. The variety X has raygenerators
u0 =
1
0
0
0
, u1 =
0
1
0
0
, u2 =
0
0
1
0
, u3 =
0
0
0
1
, u4 =
0
−1
1
0
, u5 =
2
0
−1
−1
, u6 =
−1
0
0
0
u7 =
−1
0
1
0
and the linear equivalences between the toric divisors are D0 ∼ −2D5+D6+D7, D1 ∼
D4, D2 ∼ −D4 + D5 −D7, D3 ∼ D5. Set m = 10 and let v = (x, y, z, w) ∈ P 4m. The
29
4.2. Method 1
anticanonical divisor −KX corresponds to w = (1, . . . , 1) ∈ Z8. By (4.1.4), the solution
to
qm(v,w) =
⌊ (x−1)+1m ⌋
⌊ (y−1)+1m ⌋
⌊ (z−1)+1m ⌋
⌊ (w−1)+1m ⌋
⌊ (−y+z)+1m ⌋
⌊ (2x−z−w)+1m ⌋
⌊ (−x+1)+1m ⌋
⌊ (−x+z)+1m ⌋
=
0
0
0
0
⌊−y+z+1m ⌋
⌊2x−z−w+1m ⌋
⌊−x+2m ⌋
⌊−x+z+1m ⌋
for v ∈ P 4m is an element of D(ω−1)m and similarly we can calculate D(ω−i)m by
determining qm(v, iw), for 0 ≤ i ≤ 4. It follows that |Dm| = 18, |D(ω−1)m| = 18 and|Dgen
m | = 46. For each line bundle L in the collection
L =
OX(−iD4 − jD5 − kD6), OX(−D6 −D7), 0 ≤ i, k ≤ 1
OX(−D4 −D6 −D7), OX(−D5 −D6 −D7), 0 ≤ j ≤ 2
OX(−D4 −D5 −D6 −D7)
⊂ Dm,
L−1 is nef, so L is a strong exceptional collection by Lemma 4.2. A list of rays
{ρi1 , . . . , ρij} forms a cone in Σ if and only if Di1 ∩ . . . ∩ Dij 6= ∅, so we can use
the primitive collections of X to determine which divisors do not intersect. For exam-
ple, the primitive collection {u0, u7} for X implies that D0 ∩D7 = ∅ and so we obtain
the exact sequence
0→ OX(−D0 −D7)→ OX(−D0)⊕OX(−D7)→ OX → 0.
Using the basis {[D4], [D5], [D6], [D7]} for Pic(X), rewrite the exact sequence as
0→ OX(2D5 −D6 − 2D7)→ OX(2D5 −D6 −D7)⊕OX(−D7)→ OX → 0. (4.2.2)
We can use the exact sequences determined by the primitive collections to show that
Dgenm ⊂ 〈L〉. For example, the tensor of OX(−2D5 + D7) ∈ D
genm \L with (4.2.2) gives
the exact sequence
0→ OX(−D6 −D7)→ OX(−D6)⊕OX(−2D5)→ OX(−2D5 +D7)→ 0. (4.2.3)
All of the line bundles in (4.2.3) except OX(−2D5 + D7) are in 〈L〉, hence so is
OX(−2D5 + D7). By the same method and using the exact sequences of line bundles
determined by the primitive collections for X, every line bundle in Dgenm is contained
in 〈L〉 and so L is full by Proposition 4.5.
30
CHAPTER
FIVE
QUIVER MODULI AND THE STRUCTURE SHEAF OF THE
DIAGONAL
Let X be a smooth toric variety and L = {L0, . . . , Lr} be a collection of line bundles on
X. The aim of this chapter is to introduce a map d1 of sheaves on X ×X constructed
from the line bundles in L (see (5.2.1)) and consider when the cokernel of d1 is the
structure sheaf O∆ of the diagonal embedding into X ×X; this is a necessary step in
the second method that we use to show that L generates Db(X).
In the first section we use the quiver of sections to encode the endomorphism
algebra A = End (⊕r
i=0 Li) and recall the construction of the moduli space of quiver
representations Mθ(Q,J) along with its tautological bundles. We show in the second
section that if X embeds into Mθ(Q,J) in such a way that the tautological bundles
pull back to the line bundles in L, then the cokernel of d1 is O∆ (see Proposition 5.4).
The final section in this chapter describes two different approaches to showing that this
embedding does exist, depending on whether all of the line bundles in L are nef or not
(see Propositions 5.6 and 5.8).
5.1 Quivers of Sections and Moduli Spaces of Quiver Rep-
resentations
A quiver Q consists of a vertex set Q0, an arrow set Q1 and maps h,t : Q1 → Q0
giving the vertices at the head and tail of each arrow. We assume that Q is connected,
acyclic and rooted at a unique source. A non-trivial path in Q is a sequence of arrows
p = a1 . . . ak such that h(ai) = t(ai+1) for 1 ≤ i ≤ k − 1, in which case t(p) := t(a1),
h(p) := h(ak) and supp(p) = {a1, . . . , ak}. Each vertex i ∈ Q0 gives a trivial path
ei with h(ei) = t(ei) = i. By taking the paths as a generating set and defining
multiplication to be concatenation of paths when possible and zero otherwise, we obtain
the path algebra CQ. A relation is a C-linear combination of paths in Q that share
the same head and tail and are of length at least two. If J is the two-sided ideal in CQgenerated by a finite set of relations, then we obtain the quotient algebra CQ/J and
we use the notation (Q,J) to denote the quiver with relations.
A representation W of a quiver Q assigns a C-vector space Wi to each vertex
i ∈ Q0 and a C-linear map wa : Wt(a) → Wh(a) to each arrow a ∈ Q1. The dimension
31
5.1. Quivers of Sections and Moduli Spaces of Quiver Representations
of each vector space Wi determines the dimension vector v. A morphism φ between
two representations W and V is a collection of C-linear maps φi : Wi → Vi such that
for any arrow a ∈ Q1, the following square commutes:
Wt(a)wa−−−−→ Wh(a)
φt(a)
yyφh(a)
Vt(a) −−−−→va
Vh(a)
(5.1.1)
For the quiver with relations (Q,J), we can consider representations of Q that respect
the relations in J . More precisely, a representation of (Q,J) is a representation W of
Q such that for any relation generating J , the corresponding C-linear combination of
maps between the vector spaces (Wi)i∈Q0 is set to be the zero map.
Let ZQ0 be the free abelian group of functions from Q0 to Z and ZQ1 be the free
abelian group of functions from Q1 to Z. Define Wt(Q) := {θ ∈ ZQ0 | θ(v) = 0} to
be the weight space for Q. Each θ ∈ Wt(Q) determines a stability parameter, where
a representation W is θ-(semi)stable if for every non-zero proper subrepresentation
W ′ ⊂ W we have θ(W ′) :=∑
{i|W ′
i 6=0} θi > (≥) 0. A parameter θ is generic if every θ-
semistable representation is θ-stable. A generic stability parameter θ can then be used
to construct the fine moduli space of θ-stable representations Mθ(Q), as introduced
by King [Kin94]. The space Mθ(Q) is a projective variety as Q is acyclic [Kin94,
Proposition 4.3].
Now fix the dimension vector v to be (1, . . . , 1) ∈ (ZQ0)∨, in which case Hille [Hil98,
Section 1.3] has shown that Mθ(Q) is a smooth toric variety. Note that the special
parameter ϑ := (−r, 1, 1, . . . , 1) is generic for this dimension vector, where r = |Q0|−1.
To construct Mθ(Q) explicitly, let the characteristic functions χi : Q0 → Z for i ∈ Q0
and χa : Q1 → Z for a ∈ Q1 form bases for ZQ0 and ZQ1 respectively. The incidence
map inc : ZQ1 → ZQ0 defined by inc(χa) = χh(a) − χt(a) determines the exact sequence
0 −−−−→ M −−−−→ ZQ1inc−−−−→ Wt(Q) −−−−→ 0. (5.1.2)
For a fixed generic stability parameter θ ∈ Wt(Q), let C[ya | a ∈ Q1]θ = C[NQ1 ∩
inc−1(θ)] denote the θ-graded piece. Then Mθ(Q) is the GIT quotient
Mθ(Q) = CQ1//θT = Proj
⊕
j≥0
C [ya | a ∈ Q1]jθ
(5.1.3)
where the action of T := HomZ(Wt(Q),C∗) is induced from the action of (C∗)Q0 ∼=∏i∈Q0
GL(Wi) on CQ1 determined by the incidence map. By choosing a group iso-
morphism between T and {(g0, . . . , gr) ∈ (C∗)Q0 | g0 = 1} we obtain a T -equivariant
vector bundle⊕
i∈Q0OCQ1 on CQ1 which descends to the universal family
⊕i∈Q0
Fi
on Mθ(Q) [Kin94, Proposition 5.3]. The summands Fi are called the tautological line
bundles on Mθ(Q) and F0 is the trivial line bundle, where 0 ∈ Q0 labels the source of
Q, as T acts trivially on the summand given by i = 0 in⊕
i∈Q0OCQ1 . The dimension
of Mθ(Q) is |Q1| − |Q0|+ 1 and Pic(Mθ(Q)) ∼= Wt(Q) [Hil98, Theorem 2.3].
If we are considering a quiver with ideal of relations J , we denote the fine moduli
32
5.1. Quivers of Sections and Moduli Spaces of Quiver Representations
space of θ-stable representations of Q that respect the relations in J byMθ(Q,J). By
sending a path p = a1 . . . ak to the monomial ya1 · · · yak ∈ C [ya | a ∈ Q1] and extending
linearly, we obtain a C-linear map from CQ to C [ya | a ∈ Q1]. We let IJ be the ideal
in C [ya | a ∈ Q1] generated by the image of J under this map, in which caseMθ(Q,J)
is given by the GIT quotient
Mθ(Q,J) = V(IJ)//θT = Proj
⊕
j≥0
(C [ya | a ∈ Q1] /IJ )jθ
. (5.1.4)
Let L = {L0, . . . , Lr} be a collection of non-isomorphic effective line bundles on a
projective normal toric variety X with L0 := OX . As X is projective and irreducible,
if Hom(Li, Lj) 6= 0 then Hom(Lj , Li) = 0 and so we can assume L is ordered such
that i < j whenever Hom(Li, Lj) 6= 0. The endomorphism algebra End(⊕
i Li) can be
conveniently described by its quiver of sections Q, whose vertices Q0 = {0, . . . , r} are
the line bundles in L and the number of arrows from vertex i to j for i < j is given by
the dimension of the cokernel of the map
⊕
i<k<j
Hom(Li, Lk)⊗Hom(Lk, Lj) −→ Hom(Li, Lj). (5.1.5)
A torus-invariant section s ∈ Hom(Li, Lj) is irreducible if it is not in the image of this
map. Each section in a basis of the irreducible sections determines a divisor of zeroes,
and these divisors label the arrows between vertex i and j; we therefore denote div(a)
for the divisor that labels the arrow a ∈ Q1, and div(p) :=∑
a∈supp(p) div(a) for a path
p. The corresponding labelling monomial is xdiv(p) :=∏
a∈supp(p) xdiv(a) ∈ C[xρ | ρ ∈
Σ(1)]. Note that the quiver is acyclic and as the collection is effective, the quiver is
connected and rooted at 0. The arrow labels determine the two-sided ideal of relations
J , generated by the set
{pi − pj | pi, pj paths in Q, t(pi) = t(pj),h(pi) = h(pj), div(pi) = div(pj)}. (5.1.6)
Lemma 5.1. [CS08, Proposition 3.3] Let Q be the quiver of sections for the collection
L above, with ideal of relations J . Then CQ/J ∼= End(⊕
i Li).
Each line bundle Li is isomorphic to OX(D′i) for some Cartier divisor D′
i and we con-
struct Q explicitly by computing the vertices of the polyhedron conv(NΣ(1)∩deg−1(D′i−
D′j)) for each i 6= j ∈ Q0. The vertices correspond to the torus-invariant generators of
Hom(Li, Lj), from which we pick the irreducible sections.
Example 5.2. Figure 5.1(a) displays the quiver of sections for the full strong excep-
tional collection on H1 given in Example 2.23, whilst Figure 5.1(b) lists the arrows.
The ideal of relations is
J = (a3a6 − a1a5, a3a7 − a2a5, a2a4a6 − a1a4a7).
Example 5.3. Let X be the smooth toric Fano fourfold E1 in Example 3.17 and fix
m≫ 0. Choose {[D4], [D5], [D6]} to be the basis of Pic(X); the exact sequence (2.1.3)
33
5.2. Quiver Moduli and the Structure Sheaf of the Diagonal
0 1
2 3
x0
x2
x1x3 x1
x0
x2
(a) Quiver of sections
0 1
2 3
a1
a2
a3a4 a5
a6
a7
(b) List of arrows
Figure 5.1: The quiver of sections of a full strong exceptional collection on H1
for X is
0 −−−−→ M
1 0 0 00 1 0 00 0 1 00 0 0 1−1 0 0 03 −1 −1 −12 −1 −1 −1
−−−−−−−−−−−→ Z7
[
1 0 0 0 1 0 0−3 1 1 1 0 1 0−2 1 1 1 0 0 1
]
−−−−−−−−−−−→ Pic(X) −−−−→ 0.
Every line bundle Li in the collection
L = {OX(iD5 + iD6), OX(D4 + iD5+ iD6), OX(D4 + jD5+(j+1)D6) | 0 ≤ i ≤ 3, 0 ≤ j ≤ 2}
is nef and L−1i ∈ Dm, so L is a strong exceptional collection by Lemma 4.2. The quiver
of sections Q for this collection is given in Figure 5.2.
0 1 2 3
4
5
6
7
8
9
10
x1
x2x3
x5x6
x4
x1
x2x3
x5x6
x4
x1
x2x3
x5x6
x4x0x5 x4
x0
x1x2x3
x6
x1x2x3
x5
x1x2x3
x6
x1x2x3
x5
x1x2x3
x6
x5
Figure 5.2: A quiver of sections on the smooth toric Fano fourfold E1
5.2 Quiver Moduli and the Structure Sheaf of the Diago-
nal
Let ι : ∆ → X ×X be the diagonal embedding and for two line bundles L1 and L2 on
X define
L1 ⊠ L2 := p∗1(L1)⊗ p∗2(L2)
34
5.2. Quiver Moduli and the Structure Sheaf of the Diagonal
where p1 and p2 are the projections from X ×X onto the first and second component
respectively. We define the map d1 of vector bundles on X ×X as follows. Let d1 have
domain and codomain:
d1 :⊕
a∈Q1
Lt(a) ⊠ L−1h(a) −→
⊕
i∈Q0
Li ⊠ L−1i . (5.2.1)
The summands of the vector bundles are line bundles and so are given by twists of the
Pic(X ×X)-graded module SX×X . We write SX×X = C[x1, . . . , xd, w1, . . . , wd] where
d = |ΣX(1)| to distinguish sections xi on the first copy of X in X ×X from sections
wi on the second copy. For line bundles Li and Lj on X, denote SX×X(Li, Lj) to be
a free SX×X-module generated by eLi,Ljcorresponding to the line bundle Li ⊠ Lj on
X ×X by (2.1.5). Then our map d1 sends
SX×X(Lt(a), L−1h(a)
)→ SX×X(Lh(a), L−1h(a)
)⊕ SX×X(Lt(a), L−1t(a)
)
eLt(a),L−1h(a)7→ xdiv(a)eLh(a),L
−1h(a)− wdiv(a)eLt(a),L
−1t(a).
The following proposition provides a condition as to when the cokernel of d1 is O∆.
Note that our choice of θ in the proposition will depend on our collection L, as explained
in the following section.
Proposition 5.4. Suppose that there exists a generic stability parameter θ and a closed
immersion φ : X → Mθ(Q,J) such that Li∼= φ∗(Fi) for 0 ≤ i ≤ r. Then the cokernel
of d1 in (5.2.1) is O∆.
Proof. We follow the arguments made in [Kin97]. Assume that θ and φ satisfy the
conditions in the proposition. For the opposite quiver with relations (Qop, Jop), the
stability parameter −θ is generic andMθ(Q,J) ∼=M−θ(Qop, Jop). In addition, we have
a closed immersion of X intoM−θ(Qop, Jop) such that the tautological line bundles on
M−θ(Qop, Jop) restrict to the line bundles L−1
i on X. A θ-stable representation W =
(Wi, ψa) of (Q,J) determines a (−θ)-stable representation W ∗ = (W ∗i , ψ
∗a) of (Qop, Jop)
and so for a point (x1, x2) ∈ X ×X → Mθ(Q,J) ×M−θ(Qop, Jop), the fibre over x1
parametrises the isomorphism class of a θ-stable representation V := (Vi, φa), whilst
the fibre over x2 parametrises the isomorphism class of a (−θ)-stable representation
W ∗. Therefore, the map d1 of vector bundles on X ×X from (5.2.1) restricted to the
fibre over (x1, x2) ∈ X ×X is given by
D :⊕
a∈Q1
Vt(a) ⊗W∗h(a) −→
⊕
i∈Q0
Vi ⊗W∗i .
The map D is dual to the map:
D∗ :⊕
i∈Q0
HomC(Vi,Wi) −→⊕
a∈Q1
HomC(Vt(a),Wh(a)) (5.2.2)
given by (βi) 7→ (βh(a)φa − ψaβt(a)). The kernel ker(D∗) of this map is precisely the
morphisms from V to W . As V and W are θ-stable, we have θ(V ) = θ(W ) = 0.
35
5.3. Nef And Non-Nef Collections
If f is a morphism in ker(D∗) then the image im(f) of f is a quotient of V , hence
θ(im(f)) ≤ 0. However, im(f) also injects into W implying that θ(im(f)) ≥ 0, so
θ(im(f)) = 0. Therefore, f is either an isomorphism or the zero morphism. It follows
that when W = V , the kernel of D∗ is canonically a copy of C. Using this observation,
we see that away from the diagonal of X × X the cokernel of d1 is rank zero as the
representations V and W are not isomorphic, whilst at each point on the diagonal the
cokernel restricts to a canonical copy of C. Therefore the cokernel of d1 is O∆.
5.3 Nef And Non-Nef Collections
Our choice of the generic stability parameter used in Proposition 5.4 will depend on
whether our chosen line bundles are nef or not. Firstly, assume that L is a collection
of nef line bundles on X and recall the special stability parameter ϑ = (−r, 1, 1, . . . , 1).
Craw and Smith [CS08] associate to Q a projective toric variety |L| ∼= Mϑ(Q) called
the multigraded linear series of L. They define the morphism φL : X → |L| which
factors into
X −→Mϑ(Q,J) → |L| (5.3.1)
and [CS08, Corollary 4.10] present criteria as to when φL is a closed embedding.
Lemma 5.5. [CS08, Corollary 4.10] Let L = {OX , L1, . . . , Lr} be a collection of
basepoint-free line bundles and set L =⊗
Li∈LLi. Assume that the map H0(X,L1)⊗
· · ·⊗H0(X,Lr)→ H0(X,L) is surjective. Then the morphism φL : X → |L| is a closed
embedding if and only if L is very ample.
For a line bundle L on XΣ, there is a natural bijection between deg−1(L) ∩ NΣ(1) and
the sections that generate Γ(XΣ, L). Define PL to be the polytope in RΣ(1) that is
the convex hull of the lattice points deg−1(L) ∩ NΣ(1). As the tensor operation on
sections corresponds to addition of lattice points in NΣ(1) and the Minkowski sum of
two polytopes adds each lattice point of the first polytope to every lattice point of
the second polytope, the proposition below follows immediately from Lemma 5.5 and
[CS08, Theorem 4.15]:
Proposition 5.6. Let L be a collection of nef line bundles. If L :=⊗
Li∈LLi is very
ample and the Minkowski sum of the polytopes {PLi| Li ∈ L} is equal to PL, then the
morphism φL : X → |L| is a closed embedding. In this case, we can recover the line
bundles in L as the restriction of the tautological bundle on |L| to X.
Note that as the varieties we are considering are smooth and toric, any ample line
bundle is very ample by Lemma 2.11.
If the collection L contains a line bundle that is not nef then, for the special stability
parameter ϑ, the multigraded linear series construction only gives a rational map from
X toMϑ(Q,J). In this case, we need to choose a different generic stability parameter
θ in order to obtain a closed embedding of X into Mθ(Q,J) via the map defined in
[CS08], such that the tautological bundle onMθ(Q,J) restricts to⊕
i∈Q0Li on X. To
achieve this, we recall the construction of the toric variety Yθ ⊂Mθ(Q,J) from [CS08],
(see also [CMT07] and [CQV12]).
36
5.3. Nef And Non-Nef Collections
Define the map
π := (inc,div) : ZQ1 →Wt(Q)⊕ ZΣ(1)
with image Z(Q) := π(ZQ1) and subsemigroup N(Q) := π(NQ1). The projections
π1 : Z(Q)→Wt(Q) and π2 : Z(Q)→ ZΣ(1) fit in to the commutative diagram
ZQ1
Z(Q) Wt(Q)
ZΣ(1) Pic(X)
πinc
div π1
π2 pic
deg
where pic(χi) := Li for i ∈ Q0 is a group homomorphism. Let C[N(Q)] and C[NQ1] be
the semigroup algebras defined by N(Q) and NQ1 respectively. The surjective map of
semigroup algebras π∗ : C[NQ1 ]→ C[N(Q)] induced by π has kernel IQ that defines an
affine toric subvariety V(IQ) ⊂ CQ1 . We obtain a T -action on V(IQ) via restriction of
the T -action on CQ1 . For a generic weight θ ∈Wt(Q), we have the categorical quotient
Yθ := V(IQ)//θT = Proj
⊕
j≥0
C[N(Q)]jθ
where C[N(Q)]θ is the θ-graded piece. The variety Yθ is toric and is a closed subvariety
of Mθ(Q,J).
Lemma 5.7. Let θ ∈ Wt(Q) and consider the variety Yθ. If each representation
corresponding to a torus-invariant point in Yθ is θ-stable, then θ is generic.
Proof. We follow the first part of the proof for [BCQV14, Lemma 4.2]. Let y ∈ Yθ and
Vy = ((Vy)i, φa) be the corresponding θ-semistable representation. A subrepresentation
S = (Si, sa) of Vy determines a set of vertices I ⊂ Q0 such that i ∈ I if and only
if Si = C. Let y′ be the distinguished point in the torus orbit containing y with
corresponding representation Vy′ . Then S determines a submodule S′ = (S′i, s
′a) ⊂ Vy′
where S′i = C if and only if i ∈ I. Likewise, we obtain a submodule S0 ⊂ V0 of
the module V0 corresponding to the torus-invariant point in any toric chart containing
the torus orbit, as V0 is obtained from Vy′ by setting certain maps to zero. Now
θ(S) = θ(S′) = θ(S0), but θ(S0) > 0 as V0 is assumed to be θ-stable. As this holds for
any subrepresentation of Vy, then Vy is θ-stable and hence θ is generic as y ∈ Yθ was
arbitrary.
Proposition 5.8. Fix a generic θ ∈ Wt(Q) such that L := pic(θ) is an ample line
bundle on X. If
deg−1(L) ∩NΣ(1) ⊂ π2(π−11 (θ) ∩ N(Q)
)(5.3.2)
37
5.3. Nef And Non-Nef Collections
then the homomorphism of graded rings
(π2)∗ :⊕
j≥0
C[N(Q)]jθ →⊕
j≥0
C[xρ | ρ ∈ ΣX(1)]jL
induces an isomorphism X ∼= Yθ. Furthermore, if θ and ϑ are in the same open GIT-
chamber for the T -action on V(IQ), then the tautological bundles onMθ(Q) restrict to
the line bundles Li on X.
Proof. The morphismX → Yθ is equivariant under the action of T and HomZ(Pic(X),C∗)
on V(IQ) and CΣ(1) respectively as the diagram of lattice maps
Z(Q)π1−−−−→ Wt(Q)
π2
yypic
ZΣ(1) −−−−→deg
Pic(X)
(5.3.3)
commutes, hence we obtain a rational map from X to Yθ. As L is ample, we have
X = Proj(⊕
j≥0C[xρ | ρ ∈ ΣX(1)]jL
)and so X is a closed subvariety of Yθ when the
homomorphism of graded rings
(π2)∗ :⊕
j≥0
C[N(Q)]jθ →⊕
j≥0
C[xρ | ρ ∈ ΣX(1)]jL
induced from π2 is surjective. The bundle L is very ample as X is smooth and toric, so⊕j≥0C[xρ | ρ ∈ ΣX(1)]jL is generated in the first graded piece and thus it is enough
to check surjectivity on this piece, which follows from (5.3.2). By construction we have
π2(π−11 (θ) ∩ N(Q)
)⊂ deg−1(L)∩NΣ(1) and given any two points p1, p2 ∈ π
−11 (θ)∩N(Q)
such that p1 6= p2, then π2(p1) 6= π2(p2). Therefore, (π2)∗ induces an isomorphism
X ∼= Yθ.
Now assume that θ and ϑ are in the same open GIT-chamber for the T -action on
V(IQ) and denoteMθ′(Q) byMθ′ , for θ′ ∈Wt(Q). Following the proof of [CS08, Theo-
rem 4.15], we can identify Wt(Q) with Zr by choosing (χ1−χ0, . . . , χr−χ0) to be a basis
of Wt(Q), in which case we obtain a group isomorphism between T and {(g0, . . . , gr) ∈
(C∗)Q0 | g0 = 1} ⊂ (C∗)Q0 via the projection map ZQ0 → Zr. The i-th summand in
the the T -equivariant vector bundle⊕
i∈Q0OCQ1 therefore corresponds to the SMϑ
-
module SMϑ(χi−χ0), and so the tautological line bundles onMϑ are OMϑ
,OMϑ(χ1−
χ0), . . . ,OMϑ(χr − χ0). These bundles restrict to the line bundles on V(IQ) corre-
sponding to the modules (SMϑ/IQ), (SMϑ
/IQ)(χ1 − χ0), . . . , (SMϑ/IQ)(χr − χ0). As
the chosen weight θ′ varies, the restriction of the tautological line bundles will change
if and only if the θ′-stable representations parametrised by points of V(IQ) change; as
θ and ϑ are in the same open GIT-chamber this is not the case, so they correspond
to the line bundles given by (SMθ/IQ), (SMθ
/IQ)(χ1 − χ0), . . . , (SMθ/IQ)(χr − χ0)
on Yθ. From the isomorphism X ∼= Yθ induced by (π2)∗, it follows that the module
(SMθ/IQ)(χi − χ0) corresponds to pic(χi − χ0) = Li on X.
38
CHAPTER
SIX
GENERATION OF Db(X) : RESOLUTION OF O∆ (METHOD
2)
Chapter 5 introduced a map of vector bundles d1 on X × X determined by the line
bundles in L and gave methods to determine if the cokernel of d1 isO∆. The first section
of this chapter justifies why we would want to consider this problem; we show that if d1forms part of a resolution of O∆, then our collection L generates Db(X) (see Proposition
6.1). Motivated by King (Lemmas 6.3 and 6.2), we attempt to find a resolution of O∆
by guessing a minimal projective bimodule resolution of the endomorphism algebra
of⊕
L∈L L and sheafifying the result. The second section gives the framework as to
how we guess the resolution, based on the concept of a toric cell complex introduced
by Craw–Quintero-Velez [CQV12]. The final section in this chapter brings together
this construction along with the results in Chapter 5 to present our second method for
showing that a collection of line bundles on X is full.
6.1 Resolution of O∆
Let X be a smooth projective toric variety and L = {L0, . . . , Lr} be a collection of line
bundles on X. For E ∈ Db(X × X), denote ΦE(−) := R(p1)∗(EL⊗ p∗2(−)) : Db(X) →
Db(X) to be the Fourier-Mukai transform with kernel E .
Proposition 6.1. If there exists an exact sequence of sheaves on X ×X of the form:
0→ Ek → · · · → E1d1→ E0 → O∆ → 0
where
E0 =⊕
i∈Q0
Li ⊠ L−1i ,
E1 =⊕
a∈Q1
Lt(a) ⊠ L−1h(a)
39
6.1. Resolution of O∆
and
Et =⊕
Li,Lj∈L
Lri,ti ⊠ L
−sj,tj , for 2 ≤ t ≤ k and some fixed ri,t, sj,t ∈ Z≥0,
then L classically generates Db(X).
Proof. Assume that we have a resolution of O∆ as given in the Proposition. It follows
from the projection formula that ΦO∆ is naturally isomorphic to the identity functor
on Db(X). Therefore for any object F ∈ Db(X), the object ΦO∆(F) ∼= F is classically
generated by {ΦE0(F), . . . ,ΦEk(F)}. As R(p1)∗ ◦ p∗2(−) ∼= RΓ(−) ⊗OX [Huy06, page
86] we have
ΦEt(F) ∼=⊕
Li,Lj∈L
RΓ(X,F ⊗ L−sj,tj )⊗ L
ri,ti
which is an object in 〈⊕
Li∈LLri,ti 〉 for all 0 ≤ t ≤ k. As
⊕Li∈L
Lri,ti ∈ 〈L〉 for all
0 ≤ t ≤ k, L classically generates Db(X).
In order to find a resolution of the diagonal sheaf as in Proposition 6.1, we first recall
the approach taken by King [Kin97]. For the locally free sheaf T =⊕
L∈LL−1 on X
such that HomiX(T ,T ) = 0 for i 6= 0, define A := End(T ) and T ∨ := HomOX
(T ,OX).
Note that
p∗1(T ∨) = p∗1(⊕
L∈L
L) =⊕
L∈L
p∗1(L)
and
p∗2(T ) = p∗2(⊕
L∈L
L−1) =⊕
L∈L
p∗2(L−1).
By Lemma 5.1, A is isomorphic to CQ/J for some quiver with relations (Q,J). The
following gives the final part of a minimal projective A,A-bimodule resolution of A
[Kin97].
Lemma 6.2. Let A = CQ/J and {ei | i ∈ Q0} be the indecomposable orthogonal idem-
potents. The following complex of A,A-bimodules gives the final part of the minimal
projective resolution of A.
⊕
a∈Q1
Aet(a) ⊗ [a]⊗ eh(a)A −→⊕
i∈Q0
Aei ⊗ [i]⊗ eiA (6.1.1)
where [a] and [i] are formal symbols. The map in the sequence is determined by
et(a) ⊗ [a]⊗ eh(a) 7→ a⊗ [h(a)] ⊗ eh(a) − et(a) ⊗ [t(a)]⊗ a
and the map onto A is ei ⊗ [i]⊗ ei 7→ ei.
Given a minimal projective A,A-bimodule resolution P • of A, define T ∨L
⊠A T to be
the object
p∗1(T ∨)⊗A P• ⊗A p
∗2(T ) (6.1.2)
in Db(X ×X). Using Lemma 6.2, the final map in this chain complex is the map d1from (5.2.1).
40
6.1. Resolution of O∆
Lemma 6.3. [Kin97, Theorem 1.2] If the cokernel of the map d1 in the chain complex
T ∨L
⊠A T is O∆, then T is a classical generator of Db(X).
Although the final part of the minimal projective A,A-bimodule resolution is given
by Lemma 6.2, the full resolution is not known in general and so one cannot compute
T ∨L
⊠A T . What we do instead is guess what the resolution of A is and then consider
the sheafified version of the resolution as a chain complex
S• := 0→ Sk → · · · → S2 → S1d1→ S0 (6.1.3)
of Pic(X ×X)-graded SX×X-modules, where S0 is a SX×X module corresponding to
⊕
i∈Q0
Li ⊠ L−1i
and S1 is a SX×X module corresponding to
⊕
a∈Q1
Lt(a) ⊠ L−1h(a).
If the homology groups of the chain complex S• are zero after saturation by the irrel-
evant ideal BX×X , we say that S• is exact up to saturation by BX×X , in which case it
determines an exact sequence of sheaves on X ×X by (2.1.5).
We guess the construction of S• by using the concept of the toric cell complex intro-
duced by Craw–Quintero-Velez [CQV12]. This is a combinatorial geometric structure
that encodes the minimal projective bimodule resolution for certain classes of alge-
bras; in particular, Calabi-Yau algebras in dimension 3 obtained from consistent (see
Definition 8.2) dimer models and abelian skew group algebras. Given a collection
E = {E0, . . . , Er} of rank one reflexive sheaves on a Gorenstein affine toric variety Y ,
the associated toric algebra is End(⊕r
i=0Ei). Craw–Quintero-Velez state the following
conjecture for consistent toric algebras:
Conjecture 6.4. [CQV12, Conjecture 6.4] Assume that the toric algebra associated to
E is consistent. If the global dimension of the algebra equals the dimension of Y , then
the toric cell complex exists and is constructed as in [CQV12], from which the minimal
projective bimodule resolution of the toric algebra can be recovered.
Although the endomorphism algebra of a tilting bundle T on a toric Fano variety X
is not Calabi-Yau, the endomorphism algebra of the pullback π∗(T ) on the total space
tot(ωX) of the canonical bundle is, so we guess the resolution on tot(ωX) and then
restrict it to X.
In what follows, we define a combinatorial method to guess the resolution of the
diagonal sheaf by L based on the construction in [CQV12]. Although the calculations
are lengthy and tedious, many of the steps can be achieved using a computer algorithm,
the results of which are contained in [PN15a].
41
6.2. The Toric Cell Complex
6.2 The Toric Cell Complex
For a smooth n-dimensional Fano toric variety X, set Y := tot(ωX) to be the total
space of the canonical bundle on X. A collection of line bundles L on X defines a
collection of line bundles LY on Y by pulling back along tot(ωX)→ X, and the Picard
lattice Pic(Y ) is isomorphic to Pic(X) under this map. Let Q′ be the quiver of sections
associated to LY and B = End(⊕
L∈LYL). The quiver Q′ is cyclic and naturally
embeds into Pic(Y )R. As Y is a toric variety, it has a fan Σ′ and we have the exact
sequence
0 −−−−→ M ′ −−−−→ ZΣ′(1) deg−−−−→ Pic(Y ) −−−−→ 0.
Definition 6.5. Let Q′ be the quiver above. Define Q′0 to be the set
⋃i∈Q′
0deg−1(i) ⊂
ZΣ(1) and for every arrow a ∈ Q′1 from i to j and each vertex u ∈ deg−1(i), define the
arrow a in the set Q′1 to be the arrow from u to u+ div(a) ∈ deg−1(j). The covering
quiver Q′ is the quiver in RΣ′(1) with vertex set Q′0 and arrow set Q′
1.
The embedding M ′ → ZΣ′(1) induces a projection f : RΣ′(1) → M ′R∼= Rn+1 which
restricts to f |ZΣ′(1) : ZΣ′(1) → Rn+1. This map fits into the diagram
0 −−−−→ M ′ −−−−→ ZΣ′(1) deg−−−−→ Pic(Y ) −−−−→ 0
∥∥∥yf |
ZΣ′(1)
0 −−−−→ M ′ −−−−→ Rn+1 −−−−→ Tn+1 −−−−→ 0
where Tn+1 := Rn+1/M ′ is a real (n+ 1)-torus.
If L is a full strong exceptional collection, Craw–Quintero-Velez [CQV12, Conjec-
ture 6.4] conjecture that the image of the arrows a ∈ Q′1 in Tn+1 under the map f
decomposes Tn+1 into a toric cell complex, comprising of k-cells for 0 ≤ k ≤ n + 1.
The minimal B,B-bimodule projective resolution of B that is expected to be encoded
by the toric cell complex has maps determined by differentiating k-cells with respect
to (k − 1)-cells, for 1 ≤ k ≤ n + 1. To any cell η in the toric cell complex, there is a
well-defined divisor div(η) and monomial xdiv(η) ∈ SY associated to it. By consider-
ing how the maps determined by cell differentiation produce ring homomorphisms on
SY×Y , we attempt to construct the exact sequence (6.1.3).
An anticanonical cycle in Q′ is a path p such that xdiv(p) =∏
ρ∈Σ′(1) xρ. Following
[CQV12], define the superpotential W to be the sum of all anticanonical cycles in Q′;
note that this is similar to the superpotential defined in [BSW10] but without the use
of signs, the reason for which is given in [CQV12, Section 6.3]. For two paths p and q
in Q′, the partial left derivative of p with respect to q is
∂qp :=
{r if p = rq,
0 otherwise
which can be extended by C-linearity to determine partial derivatives in CQ′. Let
P :=
{q a path in Q′ ∂qW is the sum of precisely two paths
that share neither initial nor final arrow
}
42
6.2. The Toric Cell Complex
and
J := {(p+, p−) | p± ∈ CQ′,∃q ∈ P such that ∂qW = p+ + p−}. (6.2.1)
Assume now that the dimension of X is 4. We define the following sets:
Γ′0 := Q′
0, Γ′1 := Q′
1, Γ′2 := J .
For
• (p+, p−) ∈ Γ′2, define Dp+p− := {p a path in Q′ | p is a summand in ∂p+W or
∂p−W},
• a ∈ Γ′1, define Da := {p a path in Q′ | p is a summand in ∂aW},
• i ∈ Γ′0, define Di := {p a path in Q′ | p is a summand in ∂eiW}.
Then let
Γ′3 := {Dp+p− | (p
+, p−) ∈ Γ′2}, Γ′
4 := {Da | a ∈ Γ′1}, Γ′
5 := {Di | i ∈ Γ′0}.
Remark 6.6. A set of paths P ∈ Γ′k is expected to be the 1-skeleton contained in a
k-cell in the toric cell complex for Q′, if the toric cell complex exists. The construction
of Γ′3, Γ′
4 and Γ′5 follow from the conjecture on duality between k-cells and (n−k)-cells
in [CQV12, Conjecture 6.5]. For brevity we will therefore refer to P as a k-cell.
Let P ∈ Γ′k for 1 ≤ k ≤ 5 and p ∈ P be a path. We define the head, tail and label
of P as h(P ) := h(p) ∈ Γ′0, t(P ) := t(p) ∈ Γ′
0 and div(P ) := div(p), and note that the
definitions do not depend on our choice of p. For P ′ ∈ Γ′k−i, P ∈ Γ′
k, 0 ≤ i < k ≤ 5
we write P ′ ⊂ P if for every path p ∈ P ′, there is a path q ∈ P such that p ⊂ q. If
P ′ ∈ Γ′k−1, P ∈ Γ′
k and P ′ ⊂ P , then a path q ∈ P containing a path p ∈ P ′ defines a
monomial←−∂ pq := xdiv(p
′) ∈ C[x0, . . . , xd] ∼= SY given by the label of the subpath p′ ⊂ q
from t(P ) to t(P ′), and a monomial−→∂ pq = wdiv(p′′) ∈ C[w0, . . . , wd] ∼= SY given by the
label of the subpath p′′ ⊂ q from h(P ′) to h(P ). Let RP ′,P be the set of equivalence
classes
{[(p, q)] | (p, q) ∈ P ′ × P, p ⊂ q}
where
(p, q) ∼ (p′, q′)⇔←−∂ pq =
←−∂ p′q
′.
As div(p) = div(p′) = div(P ′) and div(q) = div(q′) = div(P ), it follows that if (p, q) ∼
(p′, q′), then the label of the subpath in q from t(q) to h(p) is the same as the label
of the subpath in q′ from t(q′) to h(p′). Therefore, the label of the subpath in q from
h(p) to h(q) is the same as the label of the subpath in q′ from h(p′) to h(q′) and so
←−∂ pq =
←−∂ p′q
′ ⇔−→∂ pq =
−→∂ p′q
′.
For P ′ ∈ Γ′k−1 and P ∈ Γ′
k, define
∂P ′P :=∑
[(p,q)]∈RP ′,P
(←−∂ pq,−
−→∂ pq) ∈ SY × SY ∼= SY×Y
43
6.2. The Toric Cell Complex
if P ′ ⊂ P and 0 otherwise. The definition of ∂P ′P does not depend on the choice of
representatives for the equivalence classes in RP ′,P . We now have the maps
d′k := (∂P ′P ){P ′∈Γ′
k−1
P∈Γ′
k
} : (SY×Y )Γ′
k −→ (SY×Y )Γ′
k−1 , 0 < k ≤ 5,
eP 7→⊕
P ′∈Γ′
k−1
(∂P ′P )eP ′ .
Remark 6.7. The derivatives ∂P ′P are defined differently to how they are defined in
[CQV12] as a cell P ′ may ‘appear’ more than once in P (see Example 6.11). Conse-
quently, the property [CQV12, (4.3)] does not hold and we do not immediately obtain
an incidence function ε (see [CQV12, (4.4)]) that determines signs in the differentiations
of cells.
The construction above can be restricted to the toric Fano variety X as follows.
For any cone σ ∈ ΣX ⊂ NR, define
σ′ := Cone((0, 1), (uρ, 1) | ρ ∈ σ(1)) ⊂ NR × R.
The fan Σ′ ⊂ NR×R that has cones given by σ′ for all σ ∈ ΣX is the fan for Y [CLS11,
Proposition 7.3.1]. The toric divisors for X are in one-to-one correspondence with the
divisors for Y minus the divisor determined by the ray with generator (0, 1), which we
label ρtot. Define the subsets Γk := {P ∈ Γ′k | for all p ∈ P, xρtot ∤ xdiv(p)} ⊂ Γ′
k, for
0 ≤ k ≤ 5. Then the maps d′k restrict to
dk := (∂P ′P ){P ′∈Γk−1P∈Γk
} : (SX×X)Γk −→ (SX×X)Γk−1 , 0 < k ≤ 4, (6.2.2)
eP 7→⊕
P ′∈Γk−1
(∂P ′P )eP ′ .
The (SX×X)-modules (SX×X)Γ0 and (SX×X)Γ1 are graded as follows: for i ∈ Γ0 and
a ∈ Γ1, let SiX×X ⊂ (SX×X)Γ0 be given by SX×X(Li, L
−1i ) and Sa
X×X ⊂ (SX×X)Γ1 be
given by SX×X(Lt(a), L−1h(a)). Then (SX×X)Γ0 and (SX×X)Γ1 correspond to the bundles
⊕i∈Q0
Li ⊠ L−1i and
⊕a∈Q1
Lt(a) ⊠ L−1h(a) respectively, so the map d1 in (6.2.2) is the
map given in (5.2.1). Similarly for 2 ≤ k ≤ 4, the modules (SX×X)Γk are graded so
that they correspond to⊕
Li,Lj∈LLri,ki ⊠ L
−sj,kj , for some fixed ri,k, sj,k ∈ Z≥0. We
attempt to add signs to the terms (∂P ′P )eP ′ in the maps dk for 2 ≤ k ≤ 4 so that we
get a Pic(X ×X)-graded chain complex of SX×X -modules
0 −→ (SX×X)Γ4 d4−→ (SX×X)Γ3 d3−→ (SX×X)Γ2 d2−→ (SX×X)Γ1 d1−→ (SX×X)Γ0 (6.2.3)
In order to show that the chain complex determines an exact sequence of sheaves on
X ×X, it then needs to be checked that the chain complex is exact up to saturation
by the irrelevant ideal BX×X .
Remark 6.8. A similar construction can be obtained for a smooth toric Fano threefold
X. In this case, tot(ωX) is of dimension 4, and so the 4-cells in Γ′4 are given as the
dual cells to the 0-cells in Γ′0, the 3-cells in Γ′
3 are computed from the 1-cells in Γ′1 and
44
6.2. The Toric Cell Complex
the set of 2-cells is self-dual (see Remark 6.6).
Using the method above on the database of full strong exceptional collections of line
bundles in [PN15a], the exact sequence of sheaves S• from (6.1.3) has been computed
for all smooth toric Fano threefolds and 88 of the 124 smooth toric Fano fourfolds
[PN15a].
Example 6.9. Given the collection of line bundles L on X = H1 from Example 3.3,
the quiver of sections Q′ for the bundle π∗(⊕
L∈L L) on Y = tot(ωH1) is shown in
Figure 6.1. Vertices in the quiver with the same label are identified and the monomial
for the extra divisor is xρtot = x4. It follows that
W =a1a4a7a8 + a1a5a10 + a2a4a6a8 + a2a5a9 + a3a6a10
+ a3a7a9 + a4a7a8a1 + a4a6a8a2 + · · ·+ a9a3a7.
As Y is of dimension 3, the 0-cells are dual to the 3-cells and the 1-cells are dual to
the 2-cells, hence
Γ′0 = {0, 1, 2, 3}
Γ′1 = {a1, a2, a3, a4, a5, a6, a7, a8, a9, a10}
Γ′2 = {(a4a7a8, a5a10), (a4a6a8, a5a9), (a6a10, a7a9), (a7a8a1, a6a8a2), (a9a2, a10a1),
(a8a2a4, a10a3), (a8a1a4, a9a3), (a1a4a7, a2a4a6), (a2a5, a3a7), (a1a5, a3a6)}
Γ′3 = {(a1a4a7a8, a1a5a10, a2a4a6a8, a2a5a9, a3a6a10, a3a7a9), . . . ,
(a8a1a4a7, a8a2a4a6, a9a2a5, a9a3a7, a10a1a5, a10a3a6)}
Each arrow in {a8, a9, a10} has a label divisible by xρtot and so the sets Γk contain sets
of paths that do not traverse any of these arrows. It follows that |Γ0| = 4, |Γ1| =
7, |Γ2| = 3, |Γ3| = 0 and
d1 =
−w0 −w2 −w1 0 0 0 0
x0 x2 0 −w3 −w1 0 0
0 0 x1 x3 0 −w0 −w2
0 0 0 0 x1 x0 x2
d2 =
−w2w3 0 −w1
w0w3 −w1 0
0 w2 w0
x2w0 − x0w2 0 0
0 −x2 −x0x2x3 0 x1−x0x3 x1 0
Note that these maps correspond to those given in [Kin97, Case (ii)]. The resulting
chain complex
0 −→ (SX×X)Γ2 d2−→ (SX×X)Γ1 d1−→ (SX×X)Γ0
is exact and hence we obtain an exact sequence of sheaves
45
6.3. Method 2
0 −→ 3(L0 ⊠ L−1
3
) d2−→
2(L0 ⊠ L−1
1
)
d1−→
L0 ⊠ L−10
⊕
L1 ⊠ L−11
⊕
L2 ⊠ L−12
⊕
L3 ⊠ L−13
⊕
L0 ⊠ L−12
⊕
L1 ⊠ L−12
⊕
L1 ⊠ L−13
⊕
2(L2 ⊠ L−1
3
)
0 1
2 3
0
x0
x2
x1x3 x1
x0
x2
x1x4
x0x3x4
x2x3x4
(a) Quiver of sections
0 1
2 3
0
a1
a2
a3a4 a5
a6
a7
a8
a9
a10
(b) List of arrows
Figure 6.1: A quiver of sections on tot(ωH1)
6.3 Method 2
Given a strong exceptional collection L of line bundles on X with associated quiver Q,
we can thus proceed as follows to show that L generates Db(X):
Step 1: Using the method described in Section 6.2, construct the chain complex of Pic(X×
X)-graded SX×X -modules (6.1.3) such that St determines the sheaf Et, where
Et =⊕
Li,Lj∈L
Lri,ti ⊠ L
−sj,tj , for 2 ≤ t ≤ 4 and some fixed ri,t, sj,t ∈ Z≥0.
Check that this chain complex is exact up to saturation by BX×X .
Step 2: If L is a collection of nef line bundles then
– check that the line bundle L =⊗
Li∈LLi is ample;
– show that the Minkowski sum of the polytopes {PLi| Li ∈ L} is equal to
PL.
Then Propositions 5.6 and 5.4 imply that the exact sequence of sheaves computed
in Step 1 is a resolution of O∆, so 〈L〉 = Db(X) by Proposition 6.1.
If L contains non-nef line bundles then
46
6.3. Method 2
– choose a weight θ ∈Wt(Q) such that pic(θ) is ample and construct Yθ;
– check that θ is generic (by Lemma 5.7, it is enough to show that the rep-
resentations corresponding to each torus-invariant point of Yθ are θ-stable)
and confirm that θ and ϑ are in the same open GIT-chamber for Yθ;
– show deg−1(L) ∩ NΣ(1) ⊂ π2(π−11 (θ) ∩ N(Q)
).
Then Propositions 5.8 and 5.4 imply that the exact sequence of sheaves computed
in Step 1 is a resolution of O∆, so 〈L〉 = Db(X) by Proposition 6.1.
An example of this construction for a collection of nef line bundles on the birationally
maximal smooth toric Fano fourfold E1 is given in Example 6.11, whilst a collection
that contains non-nef line bundles on J1 is shown to be full using this method in
Example 6.12.
Example 6.10. Example 6.9 completes Step 1 of Method 2 for the strong exceptional
collection L of line bundles on H1 given in Example 3.3. Each line bundle is nef and
L =⊗
Li∈LLi = O(4D1 + 2D4) is ample. For every matrix below, each column is a
vertex of the convex polytope in RΣ(1) corresponding to Li:
L1 :
1 0
0 0
0 1
0 0
, L2 :
1 0 0
0 1 0
0 0 1
1 0 1
, L3 :
1 0 2 0
1 1 0 0
0 1 0 2
0 0 1 1
.
Adding the vertices of PL1 to PL2 , the polytope PL1 +PL2 is the convex hull of the
columns in
2 1 1 1 0 0
0 1 0 0 1 0
0 0 1 1 1 2
1 0 1 1 0 1
.
This polytope has vertices given by the columns of
2 1 0 0
0 1 1 0
0 0 1 2
1 0 0 1
.
Similarly, PL1 + PL2 + PL3 is obtained by taking the convex hull of the sum of the
vertices for PL1 + PL2 and PL3 ; the resulting polytope has vertices
4 2 0 0
0 2 2 0
0 0 2 4
2 0 0 2
.
Note that this polytope is equal to PL, so Step 2 of Method 2 is complete and hence
L is a full strong exceptional collection for H1.
47
6.3. Method 2
Example 6.11. Using the variety X and the collection of line bundles L in Example
5.3, let Y = tot(ωX) and LY be the corresponding collection of line bundles on Y . The
quiver of sections Q′ for LY is given in Figure 6.2, where vertices with the same labels
are identified. Where the arrows coincide, the labels of Q′ are given by the labels of Q
in Figure 5.2, whilst the labels of the 7 extra arrows are given in Table 6.1.
0 1 2 3
4
5
6
7
8
9
10
0
1
2
a1
a2
a3
a4
a5
a6
a7
a8
a9
a10
a11
a12
a13
a14
a16a15
a18
a17
a20
a21
a22
a19
a24
a25
a26a23
a28
a29
a30
a27
a32
a33
a34a31
a40
a36
a37
a38
a35
a39
a44
a43a42
a41
a45
a46
Figure 6.2: A quiver of sections on tot(ωX)
a t(a),h(a) div(a) 42 10,0 x0x3x7 45 9,1 x4x740 7,0 x4x7 43 10,0 x0x2x7 46 10,2 x4x6x741 10,0 x0x5x6x7 44 10,0 x0x1x7
Table 6.1: The additional arrows in a quiver of sections for tot(ωX)
The set J is
J =
(a1a7, a2a6), (a1a8, a3a6), (a1a9, a4a6), (a1a10, a5a20), (a40a1, a32a45),
(a41a1, a44a4), (a42a1, a44a3), (a43a1, a44a2), (a2a8, a3a7), (a2a9, a4a7),...
(a36a42, a38a44), (a36a43, a37a44), (a37a42, a38a43)
As |Q′0| = 11, |Q′
1| = 46 and |J | = 83, we have |Γ′0| = |Γ′
5| = 11, |Γ′1| = |Γ′
4| = 46 and
|Γ′2| = |Γ
′3| = 83. Note that P ′ = a17 ∈ Γ′
1 appears twice in P = (a11a17a25, a12a17a24) ∈
Γ′2 (see Remark 6.7). In this case, ∂P ′P = −(x1w2 + x2w1).
The monomial for the extra divisor in Y is xρtot = x7, so the sets Γk are composed
of sets of paths that do not contain any of the arrows in {a40, a41, . . . , a46}. Via this
restriction, we obtain the chain complex of Pic(X ×X)-graded (SX×X)-modules
0→ (SX×X)7d4→ (SX×X)31
d3→ (SX×X)52d2→ (SX×X)39
d1→ (SX×X)11. (6.3.1)
This complex is exact up to saturation by BX×X [PN15a] and so is an exact sequence
of sheaves on X ×X.
We now check that the cokernel of this sequence is O∆. For this collection, the
line bundle L =⊗
Li∈LLi = OX(7D4 + 15D5 + 18D6) is ample. Each column of the
matrices below is a vertex of the convex polytope in R7 for the corresponding line
48
6.3. Method 2
bundle in L:
OX(iD5 + iD6) :
0 0 0 0i 0 0 00 i 0 00 0 i 00 0 0 00 0 0 i0 0 0 i
, OX(D4 + iD5 + iD6) :
0 0 0 1 1 1 0 1i 0 0 i+2 0 0 0 00 i 0 0 i+2 0 0 00 0 i 0 0 i+2 0 01 1 1 0 0 0 1 00 0 0 1 1 1 i i+30 0 0 0 0 0 i i+2
,
OX(D4 + jD5 + (j + 1)D6) :
1 1 1 0 0 0 0 1j+3 0 0 j 0 0 0 00 j+3 0 0 j 0 0 00 0 j+3 0 0 j 0 00 0 0 1 1 1 1 00 0 0 0 0 0 j j+30 0 0 1 1 1 j+1 j+3
, i = 0, 1, 2, 3, j = 0, 1, 2.
The vertices for the polytope corresponding to L are
OX(7D4 + 15D5 + 18D6) :
3 3 3 7 7 7 0 0 0 0 724 0 0 32 0 0 15 0 0 0 00 24 0 0 32 0 0 15 0 0 00 0 24 0 0 32 0 0 15 0 04 4 4 0 0 0 7 7 7 7 00 0 0 4 4 4 0 0 0 15 360 0 0 0 0 0 3 3 3 18 32
.
Using Macaulay2 [GS, PNa] to compute the Minkowski sum of the polytopes {PLi|
Li ∈ L}, we find that it is equal to the polytope corresponding to L and so (6.3.1) is a
resolution of O∆ by Propositions 5.6 and 5.4. Therefore, L is full by Proposition 6.1.
Example 6.12. Let X be the birationally maximal smooth toric Fano fourfold J1.The primitive generators for the rays of X are
u0 =
1
0
0
0
, u1 =
0
1
0
0
, u2 =
−1
−1
−1
0
, u3 =
0
0
1
0
, u4 =
0
0
0
1
, u5 =
0
0
−1
−1
, u6 =
−1
0
0
0
, u7 =
−1
0
1
0
The collection of line bundles on X
L =
OX((2 + i)D2 + (2 + j − i)D5 + 2D6 + (k + 1)D7),
OX((1 + i)D2 + (k + i)D5 + (1 + j − i)D6 + (1 + j − i)D7),
OX(kD7), OX(D2 + kD5 +D6 +D7),
OX(3D2 +D5 + 2D6 + 2D7)
1 ≤ i ≤ j ≤ 2
0 ≤ k ≤ 1
is strong exceptional and contains the non-nef line bundle OX(D7). We obtain the
chain complex of Pic(X ×X)−graded (SX×X)−modules
0→ (SX×X)12d4→ (SX×X)38
d3→ (SX×X)59d2→ (SX×X)50
d1→ (SX×X)17 (6.3.2)
from this collection, which is exact up to saturation by BX×X [PN15a]. The quiver of
sections Q corresponding to L is shown in Figure 6.3, whilst Table 6.2 lists the labels
of its arrows.
As |Q0| = 17 and |Σ(1)| = 8, we let {ei | i ∈ Q0} ∪ {eρ | ρ ∈ Σ(1)} be the standard
basis of Z17+8 and define the lattice points ca := eh(a) − et(a) + ediv(a) for each arrow
a ∈ Q1. The map π is then given by the matrix C : Z50 → Z17+8, where the columns
of C are given by ca for a ∈ Q1 and the semigroup N(Q) is given by the lattice points
generated by positive linear combinations of the ca. Our choice of basis for Pic(X) and
49
6.3. Method 2
a t(a),h(a) div(a) 17 4,7 x3x7 34 8,12 x21 0,1 x7 18 4,9 x0 35 8,13 x32 0,2 x0 19 5,7 x1 36 9,12 x43 1,2 x1x6 20 5,7 x2 37 9,12 x54 1,2 x2x6 21 5,8 x6x7 38 9,14 x35 1,3 x3x6x7 22 5,11 x3x7 39 10,12 x76 1,4 x0x3 23 5,12 x0 40 10,13 x47 2,3 x4 24 6,8 x4 41 10,13 x58 2,3 x5 25 6,8 x5 42 10,14 x19 2,4 x3x7 26 6,9 x1 43 10,14 x210 3,4 x1 27 6,9 x2 44 11,13 x611 3,4 x2 28 6,10 x3 45 12,15 x412 3,5 x3x7 29 7,10 x6 46 12,15 x513 3,6 x0 30 7,11 x4 47 12,16 x114 4,5 x4 31 7,11 x5 48 12,16 x215 4,5 x5 32 7,16 x0 49 13,15 x716 4,6 x6x7 33 8,12 x1 50 14,16 x7
Table 6.2: The arrows in a quiver of sections for the smooth toric Fano fourfold J1
Wt(Q) imply that the lattice maps deg and pic are given by the matrices:
deg :
[1 1 1 1 0 0 0 00 0 0 1 1 1 0 01 0 0 0 0 0 1 01 0 0 −1 0 0 0 1
], pic:
[0 0 1 1 2 2 2 3 2 3 3 3 3 3 4 3 40 0 0 1 1 2 1 2 2 1 2 3 2 3 2 3 20 0 1 1 1 1 2 1 2 2 2 1 2 2 2 2 20 1 1 1 1 1 2 1 2 2 1 1 2 1 1 2 2
]
Fix θ to be the weight that assigns −6 to the vertex 0 in the quiver, 1 to the vertices
{11, 12, . . . , 16} and 0 to every other vertex. We note that pic(θ) is the ample line
bundle L = OX(20D2+15D5+11D6+9D7). For this choice of θ, π2(N(Q) ∩ (π1)−1(θ)
)
surjects onto NΣ(1) ∩ deg−1(L) [GS, PNa] and so Yθ is isomorphic to X. Any weight
θ′ such that θ′i > 0 for i > 0 is generic, for the same reason that the special parameter
ϑ is generic; any proper subrepresentation V ′ ⊆ V of a θ′-semistable representation V
has V ′0 = 0 as Q is connected and rooted at 0, hence θ′(V ′) > 0 and so V is θ′-stable.
As our choice of weight θ has θi ≥ 0 for i > 0, it is immediate that θ is in the same
closed GIT-chamber for the T -action on V(IQ) as ϑ and therefore they are in the same
open chamber if θ is generic. To check that θ is generic, it is enough to check that for
each torus-invariant point on Yθ, the corresponding representation is θ-stable. Recall
that each maximal cone corresponds to a torus-invariant point and that the point is in
the intersection of the divisors labelled by the rays of the cone – the list below gives
the 17 maximal cones in the fan for Yθ:
{ρ0, ρ1, ρ3, ρ4} {ρ0, ρ1, ρ3, ρ5} {ρ0, ρ1, ρ4, ρ5} {ρ0, ρ2, ρ3, ρ4} {ρ0, ρ2, ρ3, ρ5}
{ρ0, ρ2, ρ4, ρ5} {ρ1, ρ2, ρ4, ρ5} {ρ1, ρ2, ρ4, ρ6} {ρ1, ρ2, ρ5, ρ6} {ρ1, ρ3, ρ4, ρ7}
{ρ1, ρ3, ρ5, ρ7} {ρ1, ρ4, ρ6, ρ7} {ρ1, ρ5, ρ6, ρ7} {ρ2, ρ3, ρ4, ρ7} {ρ2, ρ3, ρ5, ρ7}
{ρ2, ρ4, ρ6, ρ7} {ρ2, ρ5, ρ6, ρ7}
For the representation (V, φ) corresponding to the torus-invariant point with rays
{ρi1 , ρi2 , ρi3 , ρi4}, the map φa is 0 if for any xi ∈ {xi1 , xi2 , xi3 , xi4}, xi divides div(a),
whilst φa = 1 otherwise. For example, consider the maximal cone {ρ0, ρ1, ρ3, ρ4}. The
50
6.3. Method 2
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
a1
a5
a16
a21
a9
a12
a17
a22
a6
a39
a49
a50
a28
a35
a38
a29
a44
a2
a3,4
a7,8
a10,11
a13
a14,15
a18
a26,27
a24,25
a19,20
a23
a30,31
a32
a36,37
a33,34
a40,41
a42,43
a47,48
a45,46
(a) The quiver of sections
0
1
2 3
4
5
6
7
8
9
10
11
12
13
14
15
16
(b) A representation of the quiver corresponding to a torus-invariant point in Yθ
Figure 6.3: A quiver of sections on the smooth toric Fano fourfold J1
corresponding representation V = (V, φ) has φa = 0 for
a ∈
{a2, a3, a5, a6, a7, a9, a10, a12, a13, a14, a17, a18, a19, a22, a23, a24, a26,
a28, a30, a32, a33, a35, a36, a38, a40, a42, a45, a47
}
and is displayed in Figure 6.3. Specifying a subrepresentation (V ′, φ′) of V is equivalent
to setting φ′a = φa for all a ∈ Q1 and choosing a subset I ⊂ Q0 so that V ′i = C for i ∈ I,
and V ′i = 0 otherwise. In our example, for any subrepresentation V ′ with V ′
0 = C, we
have V ′ = V as there is a non-zero map from V ′0 to every other V ′
i . It is also clear
from Figure 6.3 that for any i ∈ Q0, there is a non zero map from Vi to Vj for some
j ∈ {11, 12, . . . , 16}. As a result, the corresponding nonzero proper subrepresentation
V ′ of V must have V ′j = C and so θ(V ′) > 0 by the choice of θ. By considering the
subrepresentations of the representation corresponding to each of the 17 torus-invariant
51
6.3. Method 2
points on Yθ, we see that θ is generic – the calculations for this example can be found
in the file [PNa]. Therefore, (6.3.2) is a resolution of O∆ by Propositions 5.8 and 5.4,
so the collection L on J1 is full by Proposition 6.1.
52
CHAPTER
SEVEN
FULL STRONG EXCEPTIONAL COLLECTIONS ON TORIC
VARIETIES
In this chapter we present the main theorems of this thesis: we construct full strong
exceptional collections of line bundles on all smooth toric Fano fourfolds X and then
show that they pull back to give tilting bundles on tot(ωX).
We achieve the first result by choosing collections L from a certain set of line bundles
on X so that the pushforward of L onto a torus-invariant divisorial contraction X1 is
automatically full if L is full, and the pushforward coincides with the image of L
under the Picard map γ : Pic(X) → Pic(X1) (see Proposition 7.2 and Lemma 7.1).
We then check that the collection on X1 is strong exceptional computationally by
utilising the preimage of the nnnvc-cones for X1 under γ as outlined in Chapter 3;
these computations are achieved using the package QuiversToricVarieties. Using this
process, we obtain full strong exceptional collections on many of the toric Fano fourfolds
from the pushforward of collections on the birationally maximal examples (see Theorem
7.4 and Table B.3), with the rest constructed individually. To obtain tilting bundles
on tot(ωX), we show that the pullback of a full strong exceptional collection of line
bundles is tilting on tot(ωX) if it satisfies a computational condition on the vanishing
of the higher cohomology of certain line bundles on X (see Theorem 7.7); again, this
computation is performed in QuiversToricVarieties.
7.1 Tilting Bundles Comprising of Line Bundles
For a divisorial contraction (f, φ) : (X0,ΣX0) → (X1,ΣX1), the Frobenius morphism
can be used to find examples of when the pushforward of a line bundle L via f and the
image of L under the map γ from (2.1.8) are equal. Recall that for a toric variety X,
the canonical bundle is ωX = −∑
ρ∈ΣX(1)Dρ.
Lemma 7.1. Fix an integer m > 0 and let (f, φ) : (X0,ΣX0)→ (X1,ΣX1) be a torus-
equivariant extremal birational contraction between smooth n-dimensional projective
toric varieties. Let σ ⊂ ΣX0 be a maximal cone such that φ(σ) is a cone in ΣX1, and
w = 0 or w = (−1, . . . ,−1) ∈ ZΣX0(1). Then for any v ∈ Pn
m,
f∗OX0(DX0v,w,σ) = OX1(DX1
v,w,φ(σ)) (7.1.1)
53
7.1. Tilting Bundles Comprising of Line Bundles
and
D(OX1)m = {f∗LX0 | LX0 ∈ D(OX0)m}, (7.1.2)
D(ωX1)m = {f∗LX0 | LX0 ∈ D(ωX0)m}. (7.1.3)
In particular, the maps f∗ and γ coincide for OX0(DX0v,w,σ).
Proof. The result [Ueh14, Lemma 6.1] gives the case w = 0. Noting that f∗(ωX0) ∼= ωX1
[CLS11, Theorem 9.3.12], the proof can also be applied to w = (−1, . . . ,−1). The
algorithm to compute Fm(L) demonstrates the equality between f∗ and γ for the line
bundles considered.
Proposition 7.2. With the same assumptions as in Lemma 7.1, choose a collection of
line bundles L ⊂ Dm ∪D(ωX0)m. If L generates Db(X0) then the line bundles in the
image of γ(L) generate Db(X1).
Proof. Note that Rf∗OX0 = OX1 and Rf∗ωX0 = ωX1 [CLS11, Theorem 9.3.12]. We
have the equality FX1m ◦ f = f ◦ FX0
m ; indeed, consider the morphism of affine varieties
f : SpecC[Sσ] → SpecC[Sσ′ ] for two cones σ and σ′. A C[Sσ]-module N becomes a
C[Sσ′ ]-module via the operation xu · n := f#(xu) · n for xu ∈ C[Sσ′ ] and n ∈ N . It
follows that
(f# ◦ (FUσ′
m )#)(xu) · n = f#(xmu) · n = (f#(xu))m · n = ((FUσm )# ◦ f#)(xu) · n
and this globalises to give FX1m ◦ f = f ◦ FX0
m .
Consequently, Rf∗(FX0m )∗OX0 = (FX1
m )∗OX1 and Rf∗(FX0m )∗ωX0 = (FX1
m )∗ωX1 . The
result then follows by Lemmas 4.4 and 7.1.
Remark 7.3. A collection of line bundles L on X is full strong exceptional if and only
if the dual collection L−1 := {L−1 | L ∈ L} is full strong exceptional. In the following
theorem when we choose L ⊂ Dm ∪ D(ωX)m and use Method 2 to show that L is
full, we actually compute the SX×X-module chain complex using L−1, as L−1 will be
an effective collection whilst L will not be effective.
Theorem 7.4. Let X be a smooth toric Fano fourfold. There exists a full strong
exceptional collection comprising of line bundles for X. A database of these collections
can be found in [PN15a].
Proof. The algorithm to construct full strong exceptional collections of line bundles on
the smooth toric Fano fourfolds works in conjunction with Table B.3 and is as follows:
Step 1: For the fourfolds that are products of smooth toric Fano varieties of a lower di-
mension, a full strong exceptional collection of line bundles is provided by Lemma
2.24 and [Kin97, Ueh14]. This accounts for 28 of the 124 fourfolds. Beılinson’s
collection {OP4 ,OP4(1), . . . ,OP4(4)} provides a full strong exceptional collection
for P4 [Beı78].
Step 2: List every fourfold that does not have a full strong strong exceptional collection
constructed and is either
– birationally maximal, or
54
7.1. Tilting Bundles Comprising of Line Bundles
– blows up once to a fourfold that has a full strong exceptional collection
constructed.
Step 3: Set m = 10 and let X be a fourfold in the list created in Step 2. The collection
Lnef := {L ∈ Dm | L−1 is nef} ⊆ Dm is strong exceptional by Lemma 4.2. For
each X, check whether |Lnef| is equal to the number |Σ(4)| of maximal cones
in the fan for X; if this is the case, then Lnef is a candidate to be a full strong
exceptional collection. If |Lnef| < |Σ(4)| then perform a computer search using the
implementation of the nnnvc-cones in QuiversToricVarieties [PN15a, GS] to find
a strong exceptional collection L ⊆ Dm ∪ D(ωX)m such that |L| = |Σ(4)|, with
preference for collections such that L−1 is a nef collection. If no such collection can
be found, continue to search for a strong exceptional collection L not contained
in Dm ∪ D(ωX)m but with |L| = |Σ(4)|. The program QuiversToricVarieties
[PN15a] contains a database of these collections, whilst Table B.3 states how
each collection was found.
Step 4: Attempt to construct the sequence (6.2.3) for L as outlined in the first step of
Method 2 ; if this is possible, then continue with the steps in Method 2 to show
that L is full (see Chapter 6). If the sequence (6.2.3) cannot be constructed then
use Method 1 (see Proposition 4.5 and Example 4.7) to show that L is full.
The result of this step is that we have now constructed a full strong exceptional
collection of line bundles for every fourfold listed in Step 2.
Step 5: If the chosen collection L on X is contained in Dm ∪ D(ωX)m then construct
the Picard lattice maps (3.2.1) from each chain of divisorial contractions X0 :=
X → X1 → · · · → Xt (see Figure B.1). Check that {Li ⊗ L−1j | Li, Lj ∈ L}
avoids the preimages ΛI,Xkof the nnnvc-cones for 1 ≤ k ≤ t and all forbidden
sets I for Xk, as explained in Chapter 3. This process, together with the efficient
construction of the preimages ΛI,Xkgiven in Proposition 3.13, is implemented
in QuiversToricVarieties [PN15a, GS]. If {Li ⊗ L−1j | Li, Lj ∈ L} satisfies this
condition, then the collection of line bundles LXkis full strong exceptional for
each 1 ≤ k ≤ t by Lemma 3.4 and Proposition 7.2. Table B.3 details the full
strong exceptional collections obtained in this way.
Step 6: If all the smooth toric Fano fourfolds have a full strong exceptional collection of
line bundles constructed, then the algorithm finishes; otherwise, return to Step
2.
The algorithm stops after two iterations. In the first iteration, Step 1 determines full
strong exceptional collections for the 28 fourfolds that are products of lower dimensional
smooth toric Fano varieties, as well as for P4. Step 2 then lists the 26 birationally max-
imal fourfolds that do not arise as products of lower dimensional smooth toric Fano
varieties. Of these, 21 have a collection L chosen from Dm ∪D(ωX)m, whilst for the
birationally maximal variety R3 we construct a full strong exceptional collection con-
tained in Dm∪D(ωX)m from a second collection as detailed in Example 7.6. Examples
6.12 and C.2 provide more details on the construction of the resolution of the diagonal
sheaf using the non-nef collections for J1 and V 4 respectively. We then obtain full
strong exceptional collections for 64 of the fourfolds via Step 5.
55
7.1. Tilting Bundles Comprising of Line Bundles
In the second iteration, Step 2 lists the 4 non-birationally maximal fourfolds H10,
M1, M2 and M3. Steps 3 and 4 construct full strong exceptional collections for these
varieties, with Example C.1 providing more details on the construction of the resolution
of the diagonal sheaf using the non-nef collection for M1. The only remaining fourfold
without a full strong exceptional collection constructed is D16, but there is a divisorial
contraction from H10 to D16 and the collection for H10 is chosen from Dm ∪D(ωX)m.
Hence, after completing Step 5 we obtain a full strong exceptional collection for D16
from the collection on H10 and so the algorithm terminates in Step 6.
The calculations in Macaulay2 [GS] and Sage [S+15] that are required for this proof
can be found in the file [PNa].
Remark 7.5. The algorithm in the proof above can be adapted to provide a new proof
that there exist full strong exceptional collections on n-dimensional smooth toric Fano
varieties for n ≤ 3. In particular, a resolution of O∆ using the line bundles in a
collection L has been constructed for each birationally maximal Fano threefold X.
Example 7.6. The birationally maximal smooth toric Fano fourfold X := R3 has raygenerators
u0 =
1
0
0
0
, u1 =
0
1
0
0
, u2 =
0
−1
1
−1
, u3 =
0
0
1
0
, u4 =
0
0
0
1
, u5 =
0
0
−1
0
, u6 =
1
0
0
−1
,
u7 =
−1
0
0
1
, u8 =
−1
0
0
0
for its fan ΣX . We take the corresponding divisors {[D2], [D5], [D6], [D7], [D8]} to be abasis for Pic(X). The collection of nef line bundles on X
L =
OX(iD2 + jD5 +D6 +D7 +D8),
OX((j − 1)D5 + (i − 1)D7 + (i− 1)D8),
OX((i− 1)D2 + 2D5 + jD7 + jD8), OX(D2 +D5 + iD7 + jD8),
OX(D2 + 2D5 + (i − 1)D6 +D7 + jD8),
OX(2D2 + iD5 + jD7 + 2D8),
OX(2D2 + (j − i+ 1)D5 + (1− j + i)D6 +D7 + jD8)
1 ≤ i ≤ j ≤ 2
is strong exceptional and is shown to be full using Method 2, but L−1 * Dm∪D(ωX)m.However, we can use L to construct the helix HL for X and by Remark 2.28, each threadof HL is a full exceptional collection. We can therefore choose a thread in HL and twistit by the line bundle OX(−D5−D7−D8) to obtain the following full strong exceptionalcollection:
L′ =
OX(jD5 + iD7 + iD8), OX(D2 + iD7 + jD8),
OX(D2 + jD5 +D6 + iD8), OX(D2 +D5 + iD7 + jD8),
OX(2D2 + iD5 + jD7 +D8), OX(2D2 + iD5 +D6 + jD8),
OX(2D2 + (i + 1)D5 + jD6 + (i − j + 1)D8)
0 ≤ i ≤ j ≤ 1
Now (L′)−1 is a non-nef collection that is contained in Dm ∪ D(ωX)m for some m >
0, which we can use to obtain a full strong exceptional collection on the divisorial
56
7.1. Tilting Bundles Comprising of Line Bundles
contraction M4 via the method outlined in the proof of Theorem 7.4.
The full strong exceptional collections on each smooth toric Fano variety X deter-
mine tilting bundles on the total space of ωX .
Theorem 7.7. Let Y = tot(wX) be the total space of the canonical bundle on an n-
dimensional smooth toric Fano variety X, for n ≤ 4. Then Y has a tilting bundle that
decomposes as a direct sum of line bundles.
Proof. Let π : Y → X be the bundle map and L = {L0, . . . , Lr} be a full strong
exceptional collection of line bundles on X from Theorem 7.4, [Ueh14] or [Kin97]. The
collection defines a helix
HL = (. . . , L0 ⊗ ωX , . . . , Lr ⊗ ωX , L0, . . . , Lr, L0 ⊗ ω−1X , . . . , Lr ⊗ ω
−1X , . . .).
By Lemma 2.29, the pullback π∗(E) of the bundle E :=⊕r
i=0 Li is a tilting bundle on
Y if HL is geometric, which in this case is the condition that
Homk(Li ⊗ ωt1X , Lj ⊗ ω
t2X) = 0 unless k = 0
for 0 ≤ i, j ≤ r and t1 ≥ t2. This is equivalent to the condition that
Hk(X,L−1i ⊗ Lj ⊗ ω
−tX ) = 0 unless k = 0
for 0 ≤ i, j ≤ r and t ≥ 0. As ω−1X is ample, there is some positive integer T such that for
all t ≥ T , L−1i ⊗Lj⊗ω
−tX is nef for all 0 ≤ i, j ≤ r, in which case Hk(X,L−1
i ⊗Lj⊗ω−tX ) =
0 for k > 0 by Demazure vanishing [CLS11, Theorem 9.2.3]. Hence π∗(E) is a tilting
bundle if
Hk(X,L−1i ⊗ Lj ⊗ ω
−tX ) = 0 for k 6= 0, 0 ≤ i, j ≤ r, 0 ≤ t < T. (7.1.4)
The nnnvc-cones in Pic(X) can be used to show that the line bundle L−1i ⊗ Lj ⊗ ω
−tX
has vanishing higher cohomology for 0 ≤ i, j ≤ r, 0 ≤ t < T , as implemented in
QuiversToricVarieties [PN15a].
Now let X =: X0 → X1 → · · ·Xd be a chain of divisorial contractions between
smooth toric Fano varieties and assume that the collection L determines a full strong
exceptional collection on each Xk, 0 ≤ k ≤ d via the divisorial contractions, as detailed
in Theorem 7.4. For each variety Xk with collection LXk, we have an integer Tk ≥ 0
such that L−1i ⊗Lj⊗ω
−Tk
Xkis nef for all Li, Lj ∈ LXk
. Define T = max(T0, . . . , Td). Then
we can check simultaneously that each LXkdetermines a tilting bundle on tot(ωXk
) by
considering whether the line bundles L−1i ⊗Lj⊗ω
−tX avoid the preimage in Pic(X)R of all
nnnvc-cones for X1, . . . ,Xd via the Picard lattice maps, for all 0 ≤ i, j ≤ r, 0 ≤ t < T .
Again, this calculation can be performed in QuiversToricVarieties [PN15a].
Example 7.8. The full strong exceptional collection L on H1 given in Example 2.23
determines the helix
H = (. . . ,O(−D1 −D4), O, O(D1), O(D1 +D4), O(2D1 +D4), O(3D1 + 2D4), . . .).
57
7.1. Tilting Bundles Comprising of Line Bundles
The line bundles L−1s ⊗ Lt for Ls, Lt ∈ L are
{O, O(D1), O(D1 +D4), O(2D1 +D4), O(D4),
O(−D1), O(−D4), O(−D1 −D4), O(−2D1 −D4)
}
and Figure 3.1 shows them with the nnnvc-cones in Pic(H1). Using Figures 2.4 and 3.1,
we see that L−1s ⊗Lt⊗ω
−1X is nef for all Ls, Lt ∈ L; hence, following the proof of Theorem
7.7, it is enough to show that each L−1s ⊗ Lt avoids the nnnvc-cones in Pic(H1) for H
to be geometric. This immediately follows as L is a full strong exceptional collection,
so H is geometric and π∗(O⊕O(D1)⊕O(D1 +D4)⊕O(2D1 +D4)) is a tilting bundle
on tot(ωH1).
Remark 7.9. Using the full strong exceptional collections of line bundles given by King
[Kin97], Uehara [Ueh14] and Theorem 7.4, we find that the minimal T such that L−1i ⊗
Lj ⊗ ω−TX is nef for all smooth toric Fano n-folds and all collections L on the n-folds is
n− 1, for n ≤ 4. By Proposition 2.26, the endomorphism algebra of the tilting bundle
obtained on tot(ωX) from L is CY(n+ 1).
58
CHAPTER
EIGHT
FUTURE DIRECTIONS
This thesis concludes with a problem that arises when using Method 2 to show that
a strong exceptional collection of line bundles L on a smooth toric Fano variety X is
full.
Chapter 6 describes a method to construct a resolution of the diagonal sheaf O∆
from L. The construction is based on concept of the toric cell complex, introduced
by Craw–Quintero-Velez [CQV12]. They proved that the toric cell complex exists for
Calabi-Yau algebras in dimension 3 obtained from consistent dimer models and abelian
skew group algebras, and that it encodes the minimal projective bimodule resolution
of the algebra in these cases. Additionally, they conjecture that the toric cell complex
exists for consistent toric algebras obtained from varieties in higher dimensions - see
Conjecture 6.4. Using the method outlined in Chapter 6 and for the database of full
strong exceptional collections of line bundles in [PN15a], the exact sequence of sheaves
S• from (6.1.3) has been computed for all smooth toric Fano threefolds and 88 of the
124 smooth toric Fano fourfolds. These exact sequences are contained in a database
in [PN15a], and the fact that they can be computed leads us to pose the following
conjecture:
Conjecture 8.1. Let X be a smooth toric Fano threefold or one of the 88 smooth toric
Fano fourfolds such that the given full strong exceptional collection L in the database
[PN15a] has a corresponding exact sequence of sheaves S• ∈ Db(X ×X). Let B denote
the rolled up helix algebra of A = End(⊕
L∈L L−1). Then the toric cell complex of B
exists and is supported on a real four or five-dimensional torus respectively. Moreover,
• the cellular resolution exists in the sense of [CQV12], thereby producing the min-
imal projective bimodule resolution of B;
• the object S• is quasi-isomorphic to T ∨L
⊠A T ∈ Db(X×X), for T :=
⊕L∈L L
−1.
It is interesting to note that the method fails to generate a resolution of O∆ for 36
of the 124 smooth toric Fano fourfolds. An example of a problem that arises in these
cases is given in the next section.
59
8.1. The Full Strong Exceptional Collection on D1
8.1 The Full Strong Exceptional Collection on D1
We continue with the notation introduced in Chapter 5 and 6, where:
• (Q,J) was the quiver of sections encoding the endomorphism algebra of the bun-
dle⊕
L∈L L on X;
• (Q′, J ′) was the quiver of sections encoding the endomorphism algebra of the
pullback π∗(⊕
L∈L L) of this bundle to tot(ωX);
• W was the superpotential given by the sum of the anticanonical cycles in Q′.
The set J from (6.2.1) was defined by Craw–Quintero-Velez [CQV12], who use it to
generate the ideal of superpotential relations JW , an ideal in CQ′. They then generalise
the notion of consistency from dimer models as follows.
Definition 8.2. The algebra CQ′/J ′ ∼= End(π∗(⊕
L∈L L)) is consistent if J ′ and JWcoincide.
In the following example, we see that the set J does not contain enough relations
to generate J ′.
Example 8.3. The smooth toric Fano fourfold X := D1 has primitive ray generators
u0 =
1
0
0
0
, u1 =
0
1
0
0
, u2 =
−1
−1
0
2
, u3 =
0
0
1
0
, u4 =
0
0
−1
1
, u5 =
0
0
0
1
, u6 =
0
0
0
−1
and a full strong exceptional collection
L = {OX(iD0 + jD3 + kD6) | 0 ≤ i ≤ 2, 0 ≤ j, k ≤ 1}
by Theorem 7.4. Table 8.1 lists the arrows in the quiver Q′ that encodes the alge-
bra End(π∗(⊕
L∈L L)). The set of relations J is given in Figure 8.1. The relation
(a24a26a34a44, a25a50), or alternatively (a24a29a36a44, a25a50), is an example of a rela-
tion in the ideal J ′ that cannot be generated by the relations in J , and so the sequence
of sheaves (6.1.3) cannot be computed from the method outlined in Chapter 6. Note
that if p = a2a9a18, these relations appear in the derivative of the superpotential W as
∂pW = a24a26a34a44 + a24a29a36a44 + a25a50.
If we extend p to be the path p′ = a44a2a9a18a24, then we obtain the minimal relation
(a26a34, a29a36) ∈ J from ∂p′W .
As Example 8.3 shows, an important step in extending the toric cell complex to higher
dimensions is to define a set that fully contains the minimal generators of the ideal J ′.
Besides the resolutions of O∆ for the 88 smooth toric Fano fourfolds and all of
the smooth toric Fano threefolds, the smooth toric Fano fivefolds provide us with
further evidence that the toric cell complex exists in higher dimensions. There are
866 smooth toric Fano fivefolds, of which 288 are birationally maximal. Of these
60
8.1. The Full Strong Exceptional Collection on D1
a t(a),h(a) div(a)1 0,1 x0 18 3,5 x4 35 7,10 x42 0,1 x1 19 3,9 x6 36 8,10 x03 0,1 x2 20 4,5 x0 37 8,10 x14 0,2 x3 21 4,5 x1 38 8,10 x25 0,2 x4 22 4,5 x2 39 9,11 x36 0,6 x6 23 4,10 x6 40 9,11 x47 1,3 x0 24 5,6 x5 41 10,11 x08 1,3 x1 25 5,11 x6 42 10,11 x19 1,3 x2 26 6,7 x0 43 10,11 x210 1,4 x3 27 6,7 x1 44 10,0 x6x711 1,4 x4 28 6,7 x2 45 11,0 x2x4x5x712 1,7 x6 29 6,8 x3 46 11,0 x2x3x5x713 2,4 x0 30 6,8 x4 47 11,0 x1x4x5x714 2,4 x1 31 7,9 x0 48 11,0 x1x3x5x715 2,4 x2 32 7,9 x1 49 11,0 x0x4x5x716 2,8 x6 33 7,9 x2 50 11,0 x0x3x5x717 3,5 x3 34 7,10 x3 51 11,1 x6x7
Table 8.1: The arrows in a quiver of sections for tot(ωD1)
(a1a8, a2a7), (a1a9, a3a7), (a1a10, a4a13), (a1a11, a5a13), (a1a12, a6a26),(a1a44, a41a51), (a1a45, a3a49), (a1a46, a3a50), (a1a47, a2a49), (a1a48, a2a50),(a2a9, a3a8), (a2a10, a4a14), (a2a11, a5a14), (a2a12, a6a27), (a2a44, a42a51),(a2a45, a3a47), (a2a46, a3a48), (a3a10, a4a15), (a3a11, a5a15), (a3a12, a6a28),
(a3a44, a43a51), (a4a16, a6a29), (a4a35a44, a5a34a44), (a4a45, a5a46), (a4a47, a5a48),(a4a49, a5a50), (a5a16, a6a30), (a7a17, a10a20), (a7a18, a11a20), (a7a19, a12a31),
(a8a17, a10a21), (a8a18, a11a21), (a8a19, a12a32), (a9a17, a10a22), (a9a18, a11a22),(a9a19, a12a33), (a10a23, a12a34), (a10a40a51, a11a39a51), (a11a23, a12a35),
(a13a21, a14a20), (a13a22, a15a20), (a13a23, a16a36), (a14a22, a15a21), (a14a23, a16a37),(a15a23, a16a38), (a17a24a30, a18a24a29), (a17a25, a19a39), (a18a25, a19a40),
(a20a24a27, a21a24a26), (a20a24a28, a22a24a26), (a20a25, a23a41),(a21a24a28, a22a24a27), (a21a25, a23a42), (a22a25, a23a43), (a26a32, a27a31),
(a26a33, a28a31), (a26a34, a29a36), (a26a35, a30a36), (a27a33, a28a32), (a27a34, a29a37),(a27a35, a30a37), (a28a34, a29a38), (a28a35, a30a38), (a31a39, a34a41), (a31a40, a35a41),(a32a39, a34a42), (a32a40, a35a42), (a33a39, a34a43), (a33a40, a35a43), (a36a42, a37a41),(a36a43, a38a41), (a37a43, a38a42), (a39a45, a40a46), (a39a47, a40a48), (a39a49, a40a50),(a41a45, a43a49), (a41a46, a43a50), (a41a47, a42a49), (a41a48, a42a50), (a42a45, a43a47),
(a42a46, a43a48)
Figure 8.1: The set of relations J for the quiver on tot(ωD1)
birationally maximal fivefolds, 99 are such that the number of line bundles in the
collection Lnef ⊆ Dm for m≫ 0 is equal to the number of maximal cones |Σ(5)| in the
fan for that variety; these collections automatically become candidates for a full strong
exceptional collection. By adapting the method given in Chapter 6, the author has
constructed a resolution of O∆ for 26 of these varieties using Lnef. Further, it appears
that the only obstruction to providing a resolution of O∆ for the rest of the 99 fivefolds
is the lack of a sufficient set of minimal generators for the ideal J ′.
61
APPENDIX
A
QUIVERSTORICVARIETIES: A PACKAGE TO CONSTRUCT
QUIVERS OF SECTIONS ON COMPLETE TORIC VARIETIES
A.1 Introduction
For a collection of non-isomorphic line bundles L = {L0 := OX ,L1, . . . ,Lr} on a com-
plete normal toric variety X, the endomorphism algebra End(⊕
i Li) can be described
as the quotient of the path algebra of its quiver of sections by an ideal of relations
determined by labels on the arrows in the quiver [CS08]. The vertices of the quiver
correspond to the line bundles and there is a natural order on the vertices defined by
i < j if Hom(Lj , Li) = 0. For i < j, the number of arrows from i to j is equal to the
dimension of the cokernel of the map
⊕
i<k<j
Hom(Li, Lk)⊗Hom(Lk, Lj) −→ Hom(Li, Lj). (A.1.1)
A torus-invariant section s ∈ Hom(Li, Lj) is irreducible if it is not in the image of
this map. We label each arrow by the toric divisors corresponding to the sections in
a basis of the irreducible sections. Using the given order on L, the collection is strong
exceptional if
Exti(Lj , Lk) = 0,∀j, k, and i 6= 0. (A.1.2)
Let Db(X) denote the bounded derived category of coherent sheaves on X. The col-
lection L is full, or generates Db(X), if the smallest triangulated full subcategory of
Db(X) containing L is Db(X) itself. A tilting bundle T on X is a vector bundle such
that T generates Db(X) and Exti(T, T ) = 0 for i > 0; given a full strong exceptional
collection of line bundles L on X, the direct sum⊕
Li∈LLi is a tilting bundle. The
following theorem by Baer and Bondal allows us to understand Db(X) in terms of the
module category of a finite dimensional algebra.
Theorem A.1. [Bae88, Bon90] Let T be a tilting bundle on X, A = End(T ) and
Db(mod–A) be the bounded derived category of finitely generated right A-modules. Then
RHom(T,−) : Db(X)→ Db(mod–A) (A.1.3)
is an equivalence of triangulated categories.
63
A.2. Overview of the Package
A complete normal toric variety induces a short exact sequence of abelian groups
0 −−−−→ M −−−−→ ZΣ(1) deg−−−−→ Cl(X) −−−−→ 0, (A.1.4)
where M is the character lattice of the dense torus in X, Σ(1) is the set of rays in the
fan Σ of X, and the map deg sends a toric divisor D ∈ ZΣ(1) to the isomorphism class
of the rank 1 reflexive sheaf OX(D) in the class group Cl(X) (see for example [Ful93]).
Showing that L is strong exceptional in this situation is equivalent to checking that
H i(X,L−1j ⊗ Lk) = 0 for i > 0, 0 ≤ j, k ≤ r. Using a theorem of Eisenbud, Mustata
and Stillman [EMS00], we can determine if the cohomology of OX(D) vanishes by
considering when OX(D) avoids certain affine cones constructed in Cl(X), which we
call not-necessarily non-vanishing cohomology cones (nnnvc-cones) – see [PN15b]. The
purpose of the package QuiversToricVarieties for Macaulay2 [GS] is to construct the
quiver of sections for a collection of line bundles on a complete toric variety and check if
the collection is strong exceptional. We note that there does exist computer programs
that check if a collection of line bundles on a toric variety is strong exceptional; see for
example Perling’s TiltingSheaves [Per].
Restricting our attention to smooth toric Fano varieties, toric divisorial contractions
give the collection of n-dimensional toric Fano varieties a poset structure, described
for n = 3 by [Oda88] and n = 4 by [Sat00] (see also [PN15b, Remark 2.4]). The
contractions induce lattice maps between the short exact sequences (A.1.4) determined
by the varieties and these lattice maps are an essential ingredient in the proof that each
smooth toric Fano variety of dimension ≤ 4 has a full strong exceptional collection of
line bundles [PN15b, Theorem 6.4]. The package QuiversToricVarieties contains a
database of these lattice maps and of full strong exceptional collections of line bundles
on all smooth toric Fano varieties of dimension ≤ 4.
In the case whenX is a smooth toric Fano variety, let Y = tot(ωX) be the total space
of the canonical bundle on X. The package QuiversToricVarieties contains methods to
check if the pullback of a full strong exceptional collection of line bundles on X along
the morphism Y → X is a tilting bundle on Y .
QuiversToricVarieties depends on the package NormalToricVarieties for the con-
struction of toric varieties and for the database of smooth toric Fano varieties. All
varieties are defined over k = C.
A.2 Overview of the Package
Let X be a complete normal toric variety constructed in NormalToricVarieties with
a torsion-free class group. The class group lattice of X has a basis determined by
fromWDivToCl and the function fromPicToCl can be used to determine which vectors
in the lattice correspond to line bundles. The input for the method quiver is a complete
normal toric variety with a torsion-free class group, together with a list of vectors vi in
the class group lattice that correspond to the line bundles Li. The vectors are ordered
by quiver and the basis of Hom(Li, Lj) is calculated by determining the basis of the
multidegree vj−vi over the Cox ring of the variety. From this basis, the irreducible maps
are chosen and listed as arrows, with the corresponding toric divisors as labels. If some
of the vectors do not correspond to line bundles then a quiver is still constructed but
64
A.3. An Example
the resulting path algebra modulo relations may not be isomorphic to End(⊕
i∈Q0Ei),
where Ei are the rank 1 reflexive sheaves corresponding to vi. Alternatively, we can
produce a quiver by explicitly listing the vertices, the arrows with labels and the variety.
The methods source, target, label and index return the specific details of an arrow
in the quiver, a list of which can be accessed by inputting Q_1.
Besides the method quiver, the method doHigherSelfExtsVanish forms the core
of the package. The primary input is a quiver of sections. The method creates the
nnnvc-cones in the class group lattice for X and determines if the vectors vi− vj avoid
these cones. The cones are determined by certain subsets I of the rays of the fan Σ
for X; if the complement of the supporting cones for I in Σ has non-trivial reduced
homology, then I is called a forbidden set and it determines a cone in ZΣ(1). The
forbidden sets can be calculated using the function forbiddenSets, and the image of
a cone determined by a forbidden set under the map fromWDivToCl X is a nnnvc-cone
in Cl(X).
A database in NormalToricVarieties contains the smooth toric Fano varieties up
to dimension 6 and can be accessed using smoothFanoToricVariety. The divisorial
contractions between the smooth toric Fano varieties up to dimension 4 are listed under
the contractionList command, and the induced maps between their respective short
exact sequences (A.1.4) are recalled from a database in QuiversToricVarieties using
the tCharacterMap, tDivisorMap and the picardMap commands. Note that as each
variety considered is smooth, its class group is isomorphic to its Picard group.
The database containing full strong exceptional collections of line bundles for smooth
Fano toric varieties in dimension ≤ 4 can be accessed using fullStrExcColl. The
collections for the surfaces were calculated by King [Kin97], the threefolds by Costa–
Miro-Roig [CMR04], Bernardi–Tirabassi [BT09] and Uehara [Ueh14] and the fourfolds
by Prabhu-Naik [PN15b].
A.3 An Example
We illustrate the main methods in QuiversToricVarieties using the blowup of P2 at
three points, the birationally maximal smooth toric Fano surface. It is contained in the
toric Fano database in NormalToricVarieties, which is loaded by the QuiversToricVa-
rieties package.
i1 : loadPackage "QuiversToricVarieties";
i2 : X = smoothFanoToricVariety(2,4);
A full strong exceptional collection L, first considered by King [Kin97], can be recalled
from the database and its quiver of sections can be created.
i3 : L = fullStrExcColl(2,4);
o3 = {{0,0,0,0},{0,0,1,1},{0,1,0,0},{0,1,1,0},{1,0,0,0},{1,0,0,1}}
i4 : Q = quiver(L,X);
We can view the details of the quiver, either by displaying the arrows at each vertex,
or by listing all of the arrows and considering their source, target and label.
65
A.3. An Example
i5 : Q#0
o5 = HashTable{1 => {x_0x_1 , x_3x_4 } }
2 => {x_1x_2 , x_4x_5 }
3 => {x_2x_3 , x_0x_5 }
degree => {0, 0, 0, 0}
i6 : first Q_1
o6 = arrow_1
i7 : source oo, target oo, label oo
o7 = (0, 1, x_0x_1 )
The forbidden sets of rays can be computed and the collection of line bundles can be
checked to be strong exceptional. The method doHigherSelfExtsVanish creates a
copy of the nnnvc-cones in the cache table for X, where the cones are given by a vector
and a matrix {w,M} encoding the supporting closed half spaces of the cone, in which
case the lattice points of the cone are {v +w ∈ Cl(X) |MT v ≤ 0}. The non-vanishing
cone for H2 is displayed below.
i8 : peek forbiddenSets X
o8 = MutableHashTable{1 => {{0,2},{0,3},{1,3},{0,1,3},{0,2,3},{0,4},
{1,4},...}
2 => {{0,1,2,3,4,5}}
i9 : doHigherSelfExtsVanish Q
o9 = true
i10 : X.cache.cones#2
o10 = {{| -1 |, | 1 1 1 0 1 |}}
| -1 | | 1 0 1 1 1 |
| -1 | | 0 1 1 0 0 |
| -1 | | 0 0 0 1 1 |
Consider the chain of divisorial contractions X =: X4 → X3 → X2 → X0 from X
to the toric Fano surfaces numbered 3, 2 and 0 in the database. The contractions
induces lattice maps Pic(X4) → Pic(X3) → Pic(X2) → Pic(X0) and the method
doHigherExtsVanish can check if the non-isomorphic line bundles in the image of L
under these lattice maps are strong exceptional for each contraction.
i11 : doHigherSelfExtsVanish(Q,{4,3,2,0})
o11 = true
Now consider the morphism π : tot(ωX) → X. The pullback π∗(⊕
L∈LL) is a tilting
bundle on Y = tot(ωX) if
Hk(X,Li ⊗ L−1j ⊗ ω
−mX ) = 0
for all k > 0, m ≥ 0 and Li, Lj ∈ L (see for example [PN15b, Theorem 6.7]). As
ω−1X is ample, there exists a non-negative integer n such that Li ⊗ L
−1j ⊗ ω
−mX is nef
66
A.3. An Example
for 0 ≤ i, j ≤ r and m ≥ n, and hence Hk(X,Li ⊗ L−1j ⊗ ω
−mX ) = 0 for all k > 0
by Demazure vanishing. The method bundlesNefCheck checks for a given integer n
whether Li ⊗ L−1j ⊗ ω
−nX is nef for all Li, Lj ∈ L.
i12 : n=2;
i13 : bundlesNefCheck(Q,n)
o13 = true
If an integer p is included as an additional input in doHigherSelfExtsVanish, then
the method checks that for all 0 ≤ m ≤ p, whether the line bundles Li ⊗ L−1j ⊗ ω
−m
avoid the nnnvc-cones. Note that for our example, the computation above implies that
it is enough to use the integer n− 1.
i14 : doHigherSelfExtsVanish(Q,n-1)
o14 = true
For t ∈ {4, 3, 2, 0}, let {Li,t}i∈It denote the list of non-isomorphic line bundles in
the image of L under the map given by picardMap from Pic(X) → Pic(Xt), where
It is an index set. By including the list of divisorial contractions as an input in
doHigherSelfExtsVanish, we can check that
Hk(Xt, Li,t ⊗ (Lj,t)−1 ⊗ ω−m
Xt) = 0
for k > 0, 0 ≤ m ≤ n− 1, t ∈ {4, 3, 2, 0} and all i, j ∈ It.
i15 : doHigherSelfExtsVanish(Q,{4,3,2,0},n-1)
o15 = true
For all n-dimensional smooth toric Fano varieties, 1 ≤ n ≤ 3, and 88 of the 124
smooth toric Fano fourfolds, the database contains a chain complex of modules over
the Cox ring for the variety. The chain complexes are used in [PN15b] to show that
the collections of line bundles in the database for these varieties are full.
i16 : C = resOfDiag(2,4);
i17 : SS = ring C;
i18 : C
6 12 6
o18 = SS <-- SS <-- SS
67
APPENDIX
B
TABLES OF RESULTS AND THE FOURFOLD
CONTRACTION DIAGRAM
The tables in this appendix contain details on the construction of the full strong ex-
ceptional collections of line bundles L on smooth n-dimensional toric Fano varieties,
for 2 ≤ n ≤ 4. If X = Pn then the full strong exceptional collection is provided by
Beılinson [Beı78], whilst if X is a product of smooth toric Fano varieties then a full
strong exceptional collection is given for X by Lemma 2.24.
A maximal variety is a variety that is birationally maximal, as explained in Section
2.2. The sets Dm and D(ωX)m are the Frobenius pushforwards of OX and ωX respec-
tively as defined in Section 4.1, for some integer m > 0. Method 1 and Method 2
are described in Section 4.2 and Section 6.3 respectively, and are used where stated
to show that the collection L is full. The description “collection from (j)” for an n-
dimensional variety X means that there is a chain of torus-invariant divisorial contrac-
tions X0 → X1 → · · · → Xt := X between smooth n-dimensional toric Fano varieties,
where X0 is the jth n-dimensional variety. The full strong exceptional collection on X is
then given by the non-isomorphic line bundles in the image of the full strong exceptional
collection for X0 under the induced Picard lattice map γ(0→t) : Pic(X0)→ Pic(Xt) (see
Section 3.2 and Theorem 7.4 for details).
The full strong exceptional collections on the smooth toric Fano surfaces were origi-
nally described by King [Kin97]. He checked for each variety that the collections chosen
were strong exceptional, and provided resolutions of the diagonal for each collection
to show generation of Db(X). Uehara [Ueh14] observed that these collections L were
chosen from Dm, and the pushforward f∗(L) is full strong exceptional for any torus-
invariant divisorial contraction f : X1 → X2 between two smooth toric Fano surfaces
X1 and X2. Table B.1 indicates that we can use Method 2 to recover King’s resolution
of the diagonal on the maximal surface and by using Lemma 7.1, the subsequent col-
lections given for varieties (2) and (3) coincide with King’s result [Kin97] and Uehara’s
calculations [Ueh14, Theorem 6.3].
The full strong exceptional collections on the smooth toric Fano threefolds were
found by Uehara [Ueh14], having built on work by Bondal [Bon06], Costa–Miro-Roig
[CMR04] and Bernardi–Tirabassi [BT09] – for the collections L he exhibits, he shows
that f∗(L) is full strong exceptional for any torus-invariant divisorial contraction f :
X1 → X2 between two smooth toric Fano threefolds X1 and X2. We choose the
68
B.1. Toric del Pezzo Surfaces
same collections as Uehara from Dm for each maximal variety and again note that by
Lemma 7.1, our collections on the other threefolds coincide with Uehara’s collections.
Our method of generation differs from Uehara; he uses Method 1 to show that the
collections on the maximal threefolds are full, whilst we use Method 2 to find a
resolution of the diagonal.
Figure B.1 pictorially encodes the torus-invariant divisorial contractions between
the smooth toric Fano fourfolds as listed in [Sat00, Table 1] by Sato. Each line
i j
implies that there is a torus-invariant divisorial contraction from variety i to variety
j, where i and j are smooth toric Fano fourfolds listed in Table B.3. The oval nodes
indicate fourfolds that are products of lower dimensional smooth toric Fano varieties.
Each level in the diagram corresponds to the rank of the Picard lattice of a fourfold in
that level.
B.1 Toric del Pezzo Surfaces
Variety Details of the Full Strong Exceptional Collection
(0) P2 Beılinson’s collection(1) P1 × P1 product of smooth toric Fano varieties(2) S1, Bl1(P2) collection from (4)(3) S2, Bl2(P2) collection from (4)(4) S3, Bl3(P2) Maximal variety. L = Dm. Generation: Method 2
Table B.1: Tilting bundles on smooth toric Fano surfaces
B.2 Smooth Toric Fano Threefolds
Variety Details of the Full Strong Exceptional Collection
(0) P3 Beılinson’s collection
(1) B1 collection from (10)
(2) B2 collection from (17)
(3) B3 collection from (17)
(4) B4, P2 × P1 product of smooth toric Fano varieties
(5) C1 collection from (17)
(6) C2 collection from (17)
(7) C3, P1 × P1 × P1 product of smooth toric Fano varieties
(8) C4, S1 × P1 product of smooth toric Fano varieties
(9) C5 collection from (17)
(10) D1 Maximal variety. L ⊂ Dm. Generation: Method 2
(11) D2 collection from (17)
69
B.3. Smooth Toric Fano Fourfolds
(12) E1 collection from (17)
(13) E2 collection from (17)
(14) E3, S2 × P1 product of smooth toric Fano varieties
(15) E4 collection from (17)
(16) F1, S3 × P1 Maximal, product of smooth toric Fano varieties
(17) F2 Maximal variety. L = Dm. Generation: Method 2
Table B.2: Tilting bundles on smooth toric Fano threefolds
B.3 Smooth Toric Fano Fourfolds
Variety Details of the Full Strong Exceptional Collection
(0) P4 Beılinson’s collection
(1) B1 collection from (10)
(2) B2 collection from (65)
(3) B3 collection from (114)
(4) B4, P1 × P3 product of smooth toric Fano varieties
(5) B5 collection from (114)
(6) C1 collection from (100)
(7) C2 collection from (114)
(8) C3 collection from (80)
(9) C4, P2 × P2 product of smooth toric Fano varieties
(10) E1 Maximal variety. L ⊂ Dm. Generation: Method 2
(11) E2 collection from (65)
(12) E3 collection from (109)
(13) D1 collection from (100)
(14) D2 collection from (100)
(15) D3 collection from (101)
(16) D4 collection from (65)
(17) D5, P1 × B1 product of smooth toric Fano varieties
(18) D6 collection from (109)
(19) D7 collection from (115)
(20) D8 collection from (109)
(21) D9 collection from (107)
(22) D10 collection from (114)
(23) D11 collection from (114)
(24) D12, P1 × B2 product of smooth toric Fano varieties
(25) D13, P1 × P1 × P2 product of smooth toric Fano varieties
(26) D14, P1 × B3 product of smooth toric Fano varieties
(27) D15, S1 × P2 product of smooth toric Fano varieties
70
B.3. Smooth Toric Fano Fourfolds
(28) D16 collection from (47)
(29) D17 collection from (114)
(30) D18 collection from (100)
(31) D19 collection from (114)
(32) G1 collection from (81)
(33) G2 collection from (75)
(34) G3 collection from (82)
(35) G4 collection from (80)
(36) G5 collection from (82)
(37) G6 collection from (114)
(38) H1 collection from (100)
(39) H2 collection from (101)
(40) H3 collection from (100)
(41) H4 collection from (109)
(42) H5 collection from (109)
(43) H6 collection from (101)
(44) H7 collection from (100)
(45) H8, S2 × P2 product of smooth toric Fano varieties
(46) H9 collection from (109)
(47) H10 Non-maximal. L ⊂ Dm ∪D(ωX)m.
Generation: Method 2
(48) L1 collection from (108)
(49) L2 collection from (108)
(50) L3 collection from (109)
(51) L4 collection from (109)
(52) L5, P1 × C1 product of smooth toric Fano varieties
(53) L6, P1 × C2 product of smooth toric Fano varieties
(54) L7, S1 × S1 product of smooth toric Fano varieties
(55) L8, P1 × P1 × P1 × P1 product of smooth toric Fano varieties
(56) L9, S1 × P1 × P1 product of smooth toric Fano varieties
(57) L10 collection from (114)
(58) L11, P1 × C5 product of smooth toric Fano varieties
(59) L12 collection from (114)
(60) L13 collection from (115)
(61) I1 Maximal variety. L ⊂ Dm. Generation: Method 1
(62) I2 Maximal variety. L ⊂ Dm. Generation: Method 1
(63) I3 Maximal variety. L ⊂ Dm. Generation: Method 1
(64) I4 Maximal variety. L ⊂ Dm. Generation: Method 1
(65) I5 Maximal variety. L ⊂ Dm. Generation: Method 2
(66) I6 collection from (114)
(67) I7, P1 ×D1 Maximal, product of smooth toric Fano varieties
(68) I8 Maximal variety. L ⊂ Dm ∪D(ωX)m.
Generation: Method 2
(69) I9 collection from (115)
(70) I10 collection from (110)
71
B.3. Smooth Toric Fano Fourfolds
(71) I11 collection from (107)
(72) I12 collection from (114)
(73) I13, P1 ×D2 product of smooth toric Fano varieties
(74) I14 collection from (109)
(75) I15 Maximal variety. L ⊂ Dm. Generation: Method 2
(76) M1 Non-maximal. L * Dm ∪D(ωX)m.
Generation: Method 2
(77) M2 Non-maximal. L * Dm ∪D(ωX)m.
Generation: Method 1
(78) M3 Non-maximal. L ⊂ Dm ∪D(ωX)m.
Generation: Method 2
(79) M4 collection from (106)
(80) M5 Maximal variety. L ⊂ Dm ∪D(ωX)m.
Generation: Method 2
(81) J1 Maximal variety. L ⊂ Dm ∪D(ωX)m.
Generation: Method 2
(82) J2 Maximal variety. L ⊂ Dm ∪D(ωX)m.
Generation: Method 2
(83) Q1 collection from (108)
(84) Q2 collection from (109)
(85) Q3 collection from (108)
(86) Q4 collection from (110)
(87) Q5 collection from (109)
(88) Q6, P1 × E1 product of smooth toric Fano varieties
(89) Q7 collection from (114)
(90) Q8, P1 × E2 product of smooth toric Fano varieties
(91) Q9 collection from (110)
(92) Q10, S1 × S2 product of smooth toric Fano varieties
(93) Q11, P1 × P1 × S2 product of smooth toric Fano varieties
(94) Q12 collection from (114)
(95) Q13 collection from (108)
(96) Q14 collection from (109)
(97) Q15, P1 × E4 product of smooth toric Fano varieties
(98) Q16 collection from (115)
(99) Q17 collection from (114)
(100) K1 Maximal variety. L ⊂ Dm. Generation: Method 2
(101) K2 Maximal variety. L ⊂ Dm. Generation: Method 2
(102) K3 collection from (109)
(103) K4, P2 × S3 product of smooth toric Fano varieties
(104) R1 Maximal variety. L * Dm ∪D(ωX)m.
Generation: Method 2
(105) R2 Maximal variety. L * Dm ∪D(ωX)m.
Generation: Method 1
(106) R3 Maximal variety. See Example 7.6.
Generation: Method 2
72
B.3. Smooth Toric Fano Fourfolds
(107) Maximal variety. L ⊂ Dm. Generation: Method 1
(108) U1 Maximal variety. L = Dm. Generation: Method 1
(109) U2 Maximal variety. L = Dm. Generation: Method 2
(110) U3 Maximal variety. L = Dm. Generation: Method 1
(111) U4, S1 × S3 product of smooth toric Fano varieties
(112) U5, P1 × P1 × S3 product of smooth toric Fano varieties
(113) U6, P1 ×F2 Maximal, product of smooth toric Fano varieties
(114) U7 Maximal variety. L = Dm. Generation: Method 2
(115) U8 Maximal variety. L = Dm. Generation: Method 2
(116) V 4 Maximal variety. L * Dm ∪D(ωX)m.
Generation: Method 2
(117) V 4 Maximal variety. L * Dm ∪D(ωX)m.
Generation: Method 2
(118) S2 × S2 product of smooth toric Fano varieties
(119) S2 × S3 product of smooth toric Fano varieties
(120) S3 × S3 Maximal, product of smooth toric Fano varieties
(121) Z1 collection from (123)
(122) Z2 Maximal variety. L ⊂ Dm ∪D(ωX)m.
Generation: Method 1
(123) W Maximal variety. L ⊂ Dm ∪D(ωX)m.
Generation: Method 1
Table B.3: Tilting bundles on smooth toric Fano fourfolds
73
B.3.
Smooth
Toric
FanoFourfo
lds
0
1 2 3 4 5 6 7 8 9
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
38 39 40 41 42 33 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 122 121
100 101 102 103 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 107 104 105 106 123
117 112118111109 113 115108 114110
116 119
120
FigureB.1:Thetorus-in
varia
ntdiviso
rialcontra
ctionsbetw
eenthesm
ooth
toric
Fanofour-
folds
74
APPENDIX
C
FURTHER EXAMPLES
The two examples below prove that the non-nef collections of line bundles on the
smooth toric Fano fourfolds M1 and V 4 are full strong exceptional, using Method 2.
Example C.1. Let X be the smooth toric Fano fourfold M1. The primitive generatorsfor the rays of X are
u0 =
1
0
0
0
, u1 =
0
1
0
0
, u2 =
−1
−1
1
1
, u3 =
0
0
1
0
, u4 =
0
0
−1
0
, u5 =
0
0
0
1
, u6 =
0
0
0
−1
, u7 =
−1
0
0
0
The collection of line bundles on X
L =
OX(kD2 + iD6 + jD7),OX (jD2 +D4 + iD6 +D7),
OX((k − 1)D2 + (j − 1)D4 + (1 + i− j)D6),
OX((k + 1)D4 + (k + 1)D6 + (k + 1)D7)
0 ≤ i ≤ j ≤ 1
0 ≤ k ≤ 1
is strong exceptional and contains the non-nef line bundles {OX(−D2+D6),OX (−D2+
D4), OX(−D2 +D4 +D6),OX(D7),OX (D6 +D7),OX(D4 +D7),OX(D2)}. We obtain
the chain complex of Pic(X ×X)−graded (SX×X)−modules
0→ (SX×X)10d4→ (SX×X)43
d3→ (SX×X)76d2→ (SX×X)60
d1→ (SX×X)17
from this collection, which is exact up to saturation by BX×X [PN15a, GS]. Table C.1
lists the arrows in the quiver of sections Q corresponding to L.
As |Q0| = 17 and |Σ(1)| = 8, we let {ei | i ∈ Q0} ∪ {eρ | ρ ∈ Σ(1)} be the standard
basis of Z17+8 and define the lattice points ca := eh(a) − et(a) + ediv(a) for each arrow
a ∈ Q1. The map π is then given by the matrix C : Z60 → Z17+8 where the columns
of C are given by ca, a ∈ Q1, and the semigroup N(Q) is given by the lattice points
generated by positive linear combinations of the ca. Our choice of bases for Pic(X)
and Wt(Q) imply that the lattice maps deg and pic are given by the matrices:
deg :
[1 1 1 −1 0 −1 0 00 0 0 1 1 0 0 00 0 0 0 0 1 1 01 0 0 0 0 0 0 1
], pic:
[0 −1 −1 0 1 0 0 1 −1 0 0 0 1 1 0 1 00 0 1 0 0 0 1 0 1 0 1 1 0 1 1 1 20 1 0 0 0 1 0 0 1 1 0 1 1 0 1 1 20 0 0 1 0 0 0 1 0 1 1 0 1 1 1 1 2
]
Fix θ to be the weight that assigns −2 to the vertex 0 in the quiver, 1 to the vertices
75
a t(a),h(a) div(a)1 0,1 x5 21 4,5 x5 41 9,14 x42 0,2 x3 22 4,6 x3 42 10,13 x13 0,3 x7 23 4,7 x7 43 10,13 x24 0,4 x1 24 5,8 x3 44 10,14 x65 0,4 x2 25 5,9 x7 45 11,14 x76 0,5 x6 26 5,11 x4 46 11,15 x07 0,6 x4 27 5,12 x0 47 12,14 x38 0,7 x0 28 6,8 x5 48 12,15 x49 1,5 x1 29 6,10 x7 49 13,14 x510 1,5 x2 30 6,11 x6 50 13,15 x611 1,8 x4 31 6,13 x0 51 14,15 x112 1,9 x0 32 7,9 x5 52 14,15 x213 2,6 x1 33 7,10 x3 53 14,16 x0x4x514 2,6 x2 34 7,12 x6 54 14,16 x0x3x615 2,8 x6 35 7,13 x4 55 14,16 x4x6x716 2,10 x0 36 8,11 x1 56 15,16 x0x3x517 3,7 x1 37 8,11 x2 57 15,16 x1x3x5x718 3,7 x2 38 8,14 x0 58 15,16 x2x3x5x719 3,9 x6 39 9,12 x1 59 15,16 x4x5x720 3,10 x4 40 9,12 x2 60 15,16 x3x6x7
Table C.1: The arrows in a quiver of sections for the smooth toric Fano fourfold M1
15 and 16 and 0 to every other vertex. We note that pic(θ) is the ample line bundle
L = OX(D2 + 3D4 + 3D6 + 3D7). For this choice of θ, π2(N(Q) ∩ (π1)−1(θ)
)surjects
onto NΣ(1) ∩ deg−1(L) and so Yθ is isomorphic to X. As θi ≥ 0 for i > 0, θ is in the
same closed GIT-chamber for the T -action on V(IQ) as ϑ and so they are in the same
open chamber if θ is generic. To check that θ is generic, it is enough to check that for
each torus-invariant point on Yθ, the corresponding representation is θ-stable. There
are 17 maximal cones in the fan for Yθ – recall that each maximal cone corresponds to
a torus-invariant point.
For each quiver that describes a torus-invariant representation in Yθ, we need to
specify a path from the vertex 0 to vertices 15 and 16, and a path from every other
vertex to the vertex 15 or 16. Examples of these paths are given in Table C.2 and are
computed in [PNa]. As a result, every torus-invariant θ-semistable representation of Q
is θ-stable, so θ is generic and the collection L on X is full by Propositions 5.8, 5.4 and
6.1.
Example C.2. Let X be the smooth toric Fano fourfold V 4. The primitive generatorsare
u0 =
1
0
0
0
, u1 =
−1
0
0
0
, u2 =
0
1
0
0
, u3 =
0
−1
0
0
, u4 =
0
0
1
0
, u5 =
0
0
−1
0
, u6 =
0
0
0
1
, u7 =
0
0
0
−1
,
u8 =
1
1
1
1
The strong exceptional collection of line bundles L is given by the columns of the
76
Torus-Invariant (0→ 15, via {i1, . . . , ij1}), (vertex i, i→ 15 or i→ 16,Point (0→ 16, via {i1, . . . , ij2}) via vertices {i1, . . . , ij3})
{ρ0, ρ1, ρ3, ρ5} (a3a18a34a48, {3, 7, 12}), (1, a10a25a40a48, {5, 9, 12}),(a3a19a41a55, {3, 9, 14}) (2, a14a29a43a50, {6, 10, 13}),
(4, a23a34a48, {7, 12}), (8, a37a45a52, {11, 14}){ρ0, ρ1, ρ3, ρ6} (a1a10a25a40a48a59, (2, a14a28a37a45a52, {6, 8, 11, 14}),
{1, 5, 9, 12, 15}) (3, a18a32a40a48, {7, 9, 12}),(4, a21a25a40a48, {5, 9, 12}),
(10, a43a49a52, {13, 14}){ρ0, ρ1, ρ4, ρ5} (0, a2a14a29a43a50a60, (1, a10a24a37a45a52, {5, 8, 11, 14}),
{2, 6, 10, 13, 15}) (3, a18a33a43a50, {7, 10, 13}),(4, a22a29a43a50, {6, 10, 13}),
(9, a40a47a52, {12, 14}){ρ0, ρ1, ρ4, ρ6} (0, a1a10a24a37a45a52a58, (2, a14a29a43a49a52, {6, 10, 13, 14}),
{1, 5, 8, 11, 14, 15}) (3, a18a32a40a47a52, {7, 9, 12, 14}),(4, a21a24a37a45a52), {5, 8, 11, 14})
{ρ0, ρ2, ρ3, ρ5} (0, a3a17a34a48, {3, 7, 12}), (1, a9a25a39a48, {5, 9, 12}),(0, a3a19a41a55, {3, 9, 14}) (2, a13a29a42a50, {6, 10, 13}),
(4, a23a34a48, {7, 12}), (8, a36a45a51, {11, 14})...
......
{ρ2, ρ4, ρ6, ρ7} (0, a1a9a24a36a46a56, (2, a13a28a36a46, {6, 8, 11}),{1, 5, 8, 11, 15}) (3, a17a32a39a47a51, {7, 9, 12, 14}),
(4, a21a24a36a46, {5, 8, 11}),(10, a42a49a51, {13, 14})
Table C.2: Paths in the quiver associated to each torus-invariant representation in Yθ ∼= M1
matrix pic below, where we choose the divisors {[D1], [D3], [D4], [D6], [D8]} as a basis
of Pic(X). This collection contains the non-nef line bundle OX(D8). We obtain the
chain complex of Pic(X ×X)−graded (SX×X)−modules
0→ (SX×X)18d4→ (SX×X)78
d3→ (SX×X)124d2→ (SX×X)87
d1→ (SX×X)23
from this collection, which is exact up to saturation by BX×X [PN15a, GS]. The lattice
maps deg and pic are given by the matrices:
deg :
[1 1 0 0 0 0 0 0 00 0 1 1 0 0 0 0 00 0 0 0 1 1 0 0 00 0 0 0 0 0 1 1 0−1 0 −1 0 0 1 0 1 1
], pic:
[0 0 1 1 1 1 1 2 1 1 1 2 2 2 1 2 2 2 1 2 2 2 20 0 1 1 1 1 2 1 1 2 2 1 1 2 2 1 2 2 2 1 2 2 20 0 1 1 1 2 1 1 2 1 2 1 2 1 2 2 1 2 2 2 1 2 20 0 1 1 2 1 1 1 2 2 1 2 1 1 2 2 2 1 2 2 2 1 20 1 0 1 1 1 0 0 2 1 1 1 1 0 1 1 0 0 2 2 1 1 0
]
Fix θ to be the weight that assigns −9 to vertex 0, 1 to vertices {14, 15, . . . , 22} and
0 to all other vertices. We note that pic(θ) is the ample line bundle L = OX(16D1 +
16D3 + 16D4 + 16D6 + 8D8). For this choice of θ, π2(N(Q) ∩ (π1)
−1(θ))
surjects onto
NΣ(1) ∩ deg−1(L) and so Yθ is isomorphic to X. As θi ≥ 0 for i > 0, θ is in the same
closed GIT-chamber for the T -action on V(IQ) as ϑ and so they are in the same open
chamber if θ is generic. For each quiver that describes a torus-invariant representation
in Yθ, we need to specify paths from the vertex 0 to the vertices {14, 15, . . . , 22}, and
a path from every other vertex to one of the vertices in {14, 15, . . . , 22} to show that θ
is generic. These paths, as well as the other necessary computations for this example
are found in [PNa]. As a result, every torus-invariant θ-semistable representation of Q
is θ-stable, so θ is generic and the collection L on X is full by Propositions 5.8, 5.4 and
77
6.1.
78
BIBLIOGRAPHY
[Bae88] Dagmar Baer. Tilting sheaves in representation theory of algebras.
Manuscripta Math., 60(3):323–347, 1988.
[Bat82a] V. V. Batyrev. Boundedness of the degree of multidimensional toric Fano
varieties. Vestnik Moskov. Univ. Ser. I Mat. Mekh., (1):22–27, 76–77, 1982.
[Bat82b] V. V. Batyrev. Toroidal fano 3-folds. Math. USSR-Izv., 19:13–25, 1982.
[Bat99] V. V. Batyrev. On the classification of toric Fano 4-folds. J. Math. Sci. (New
York), 94(1):1021–1050, 1999. Algebraic geometry, 9.
[BCQV14] Raf Bocklandt, Alastair Craw, and Alexander Quintero Velez. Geometric
reid’s recipe for dimer models. Mathematische Annalen, pages 1–35, 2014.
[Beı78] A. A. Beılinson. Coherent sheaves on Pn and problems in linear algebra. Funk-
tsional. Anal. i Prilozhen., 12(3):68–69, 1978.
[Bon90] A. I. Bondal. Helices, representations of quivers and Koszul algebras. In Helices
and vector bundles, volume 148 of London Math. Soc. Lecture Note Ser., pages
75–95. Cambridge Univ. Press, Cambridge, 1990.
[Bon06] Alexey Bondal. Derived categories of toric varieties. In Convex and Algebraic
geometry, Oberwolfach conference reports, EMS Publishing House, volume 3,
pages 284–286, 2006.
[BS10] Tom Bridgeland and David Stern. Helices on del Pezzo surfaces and tilting
Calabi-Yau algebras. Adv. Math., 224(4):1672–1716, 2010.
[BSW10] Raf Bocklandt, Travis Schedler, and Michael Wemyss. Superpotentials and
higher order derivations. J. Pure Appl. Algebra, 214(9):1501–1522, 2010.
[BT09] Alessandro Bernardi and Sofia Tirabassi. Derived categories of toric Fano
3-folds via the Frobenius morphism. Matematiche (Catania), 64(2):117–154,
2009.
[CLS11] David A. Cox, John B. Little, and Henry K. Schenck. Toric varieties, vol-
ume 124 of Graduate Studies in Mathematics. American Mathematical Society,
Providence, RI, 2011.
80
Bibliography
[CMR04] L. Costa and R. M. Miro-Roig. Tilting sheaves on toric varieties. Math. Z.,
248(4):849–865, 2004.
[CMT07] Alastair Craw, Diane Maclagan, and Rekha R. Thomas. Moduli of McKay
quiver representations. I. The coherent component. Proc. Lond. Math. Soc. (3),
95(1):179–198, 2007.
[Cox95] David A. Cox. The homogeneous coordinate ring of a toric variety. J. Algebraic
Geom., 4(1):17–50, 1995.
[CQV12] Alastair Craw and Alexander Quintero Velez. Cellular resolutions of noncom-
mutative toric algebras from superpotentials. Adv. Math., 229(3):1516–1554,
2012.
[CS08] Alastair Craw and Gregory G. Smith. Projective toric varieties as fine moduli
spaces of quiver representations. Amer. J. Math., 130(6):1509–1534, 2008.
[DLM09] Arijit Dey, Micha l Lason, and Mateusz Micha lek. Derived category of toric
varieties with Picard number three. Matematiche (Catania), 64(2):99–116,
2009.
[Efi10] Alexander I. Efimov. Maximal lengths of exceptional collections of line bundles.
arXiv:1010.3755v2, 2010.
[EMS00] David Eisenbud, Mircea Mustata, and Mike Stillman. Cohomology on toric
varieties and local cohomology with monomial supports. J. Symbolic Com-
put., 29(4-5):583–600, 2000. Symbolic computation in algebra, analysis, and
geometry (Berkeley, CA, 1998).
[Ful93] William Fulton. Introduction to toric varieties, volume 131 of Annals of Math-
ematics Studies. Princeton University Press, Princeton, NJ, 1993. The William
H. Roever Lectures in Geometry.
[Gin06] V. Ginzburg. Calabi-Yau algebras. arXiv:math/0612139v3, 2006.
[GS] Daniel R. Grayson and Michael E. Stillman. Macaulay2, a software system
for research in algebraic geometry. Available at http://www.math.uiuc.edu/
Macaulay2/.
[Hil98] Lutz Hille. Toric quiver varieties. In Algebras and modules, II (Geiranger,
1996), volume 24 of CMS Conf. Proc., pages 311–325. Amer. Math. Soc., Prov-
idence, RI, 1998.
[HP06] Lutz Hille and Markus Perling. A counterexample to King’s conjecture. Com-
pos. Math., 142(6):1507–1521, 2006.
[Huy06] D. Huybrechts. Fourier-Mukai transforms in algebraic geometry. Oxford
Mathematical Monographs. The Clarendon Press, Oxford University Press, Ox-
ford, 2006.
[Kaw06] Yujiro Kawamata. Derived categories of toric varieties. Michigan Math. J.,
54(3):517–535, 2006.
81
Bibliography
[Kin94] A. D. King. Moduli of representations of finite-dimensional algebras. Quart.
J. Math. Oxford Ser. (2), 45(180):515–530, 1994.
[Kin97] Alastair King. Tilting bundles on some rational surfaces. Available as a
preprint at http://people.bath.ac.uk/masadk/papers, 1997.
[KN09] Maximilian Kreuzer and Benjamin Nill. Classification of toric Fano 5-folds.
Adv. Geom., 9(1):85–97, 2009.
[LM11] Micha l Lason and Mateusz Micha lek. On the full, strongly exceptional collec-
tions on toric varieties with Picard number three. Collect. Math., 62(3):275–296,
2011.
[Øb07] M. Øbro. An algorithm for the classification of smooth fano polytopes. arXiv:
0704.0049, 2007.
[Oda88] Tadao Oda. Convex bodies and algebraic geometry, volume 15 of Ergebnisse
der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related
Areas (3)]. Springer-Verlag, Berlin, 1988. An introduction to the theory of toric
varieties, Translated from the Japanese.
[Per] Markus Perling. Tiltingsheaves, a program for computing exceptional se-
quences on toric varieties. Available at http://www.mathematik.uni-kl.de/
~perling/ts/ts.html.
[PNa] Nathan Prabhu-Naik. Generationtoricfanofourfolds - a file for
macaulay2. Available at http://people.bath.ac.uk/ac886/students/
nathanPrabhuNaik.
[PNb] Nathan Prabhu-Naik. Strongexccollsmaxfourfolds - a file for macaulay2. Avail-
able at http://people.bath.ac.uk/ac886/students/nathanPrabhuNaik.
[PN15a] Nathan Prabhu-Naik. Quiverstoricvarieties: A package to construct quivers
of sections on complete toric varieties. arXiv:1501.05861v2, 2015.
[PN15b] Nathan Prabhu-Naik. Tilting bundles on toric fano fourfolds. arXiv:1501.
05871v2, 2015.
[S+15] W. A. Stein et al. Sage Mathematics Software (Version 6.1.1). The Sage De-
velopment Team, 2015. http://www.sagemath.org.
[Sat00] Hiroshi Sato. Toward the classification of higher-dimensional toric Fano vari-
eties. Tohoku Math. J. (2), 52(3):383–413, 2000.
[Tho00] Jesper Funch Thomsen. Frobenius direct images of line bundles on toric vari-
eties. J. Algebra, 226(2):865–874, 2000.
[Ueh14] Hokuto Uehara. Exceptional collections on toric Fano threefolds and birational
geometry. Internat. J. Math., 25(7):1450072, 32, 2014.
[VdB04] Michel Van den Bergh. Three-dimensional flops and noncommutative rings.
Duke Math. J., 122(3):423–455, 2004.
82
Bibliography
[WW82] Keiichi Watanabe and Masayuki Watanabe. The classification of Fano 3-folds
with torus embeddings. Tokyo J. Math., 5(1):37–48, 1982.
83